Add app

.gitignore

@@ -0,0 +1 @@
+**/__pycache__/

app.py

@@ -0,0 +1,330 @@
+import gradio as gr
+from translator import Translator
+
+history_list = [['', '', '']]
+
+translator = Translator()
+
+def translate(data):
+    if data[in_selected] == 'Tiếng Việt':
+        sen = data[vie_in]
+        src_lang = 'vi'
+    elif data[in_selected] == 'Tiếng Anh':
+        sen = data[eng_in]
+        src_lang = 'en'
+    elif data[in_selected] == 'Tiếng Lào':
+        sen = data[lao_in]
+        src_lang = 'lo'
+    elif data[in_selected] == 'Tiếng Trung':
+        sen = data[chi_in]
+        src_lang = 'zh'
+    elif data[in_selected] == 'Tiếng Khơ-me':
+        sen = data[kho_in]
+        src_lang = 'km'
+
+    if data[out_selected] == 'Tiếng Việt':
+        tgt_lang = 'vi'
+        return {vie_out: translator.translate(sen, src_lang, tgt_lang, data[models])}
+    if data[out_selected] == 'Tiếng Anh':
+        tgt_lang = 'en'
+        return {eng_out: translator.translate(sen, src_lang, tgt_lang, data[models])}
+    if data[out_selected] == 'Tiếng Lào':
+        tgt_lang = 'lo'
+        return {lao_out: translator.translate(sen, src_lang, tgt_lang, data[models])}
+    if data[out_selected] == 'Tiếng Trung':
+        tgt_lang = 'zh'
+        return {chi_out: translator.translate(sen, src_lang, tgt_lang, data[models])}
+    tgt_lang = 'km'
+    return {kho_out: translator.translate(sen, src_lang, tgt_lang, data[models])}
+
+def clear_all_func():
+    return {vie_in: '', eng_in: '', lao_in: '', chi_in: '', kho_in: '', vie_out: '', eng_out: '', lao_out: '', chi_out: '', kho_out: ''}
+
+def default_func():
+    return {vie_in: '', eng_in: '', lao_in: '', chi_in: '', kho_in: '', models: 'Baseline'}
+
+def set_input(data):
+    if data[lang] == 'Tiếng Việt':
+        return {vie_in: data[sentence]}
+    elif data[lang] == 'Tiếng Anh':
+        return {eng_in: data[sentence]}
+    elif data[lang] == 'Tiếng Lào':
+        return {lao_in: data[sentence]}
+    elif data[lang] == 'Tiếng Trung':
+        return {chi_in: data[sentence]}
+    elif data[lang] == 'Tiếng Khơ-me':
+        return {kho_in: data[sentence]}
+    return {}
+
+def add_history(data):
+    if data[in_selected] == 'Tiếng Việt':
+        src = data[vie_in]
+    elif data[in_selected] == 'Tiếng Anh':
+        src = data[eng_in]
+    elif data[in_selected] == 'Tiếng Lào':
+        src = data[lao_in]
+    elif data[in_selected] == 'Tiếng Trung':
+        src = data[chi_in]
+    elif data[in_selected] == 'Tiếng Khơ-me':
+        src = data[kho_in]
+
+    if data[out_selected] == 'Tiếng Việt':
+        tgt = data[vie_out]
+    elif data[out_selected] == 'Tiếng Anh':
+        tgt = data[eng_out]
+    elif data[out_selected] == 'Tiếng Lào':
+        tgt = data[lao_out]
+    elif data[out_selected] == 'Tiếng Trung':
+        tgt = data[chi_out]
+    elif data[out_selected] == 'Tiếng Khơ-me':
+        tgt = data[kho_out]
+
+    if src == '':
+        if len(history_list) == 1:
+            return gr.Dataset(samples=history_list)
+        else:
+            return gr.Dataset(samples=history_list[1:])
+
+    choice = data[models]
+    history_list.append([src, tgt, choice])
+    return gr.Dataset(samples=history_list[1:])
+
+def clear_all_history_func():
+    global history_list
+    history_list = [['', '', '']]
+    return gr.Dataset(samples=history_list)
+
+def change_input_tab(data):
+    if vie_in in data.keys():
+        return 'Tiếng Việt'
+    if eng_in in data.keys():
+        return 'Tiếng Anh'
+    if lao_in in data.keys():
+        return 'Tiếng Lào'
+    if chi_in in data.keys():
+        return 'Tiếng Trung'
+    return 'Tiếng Khơ-me'
+
+def change_output_tab(data):
+    if vie_out in data.keys():
+        return 'Tiếng Việt'
+    if eng_out in data.keys():
+        return 'Tiếng Anh'
+    if lao_out in data.keys():
+        return 'Tiếng Lào'
+    if chi_out in data.keys():
+        return 'Tiếng Trung'
+    return 'Tiếng Khơ-me'
+
+js = """
+function createGradioAnimation() {
+    var container = document.createElement('div');
+    container.id = 'gradio-animation';
+    container.style.fontSize = '2em';
+    container.style.fontWeight = 'bold';
+    container.style.textAlign = 'center';
+    container.style.marginBottom = '20px';
+
+    var text = 'Dịch Song Ngữ';
+    for (var i = 0; i < text.length; i++) {
+        (function(i){
+            setTimeout(function(){
+                var letter = document.createElement('span');
+                letter.style.opacity = '0';
+                letter.style.transition = 'opacity 0.5s';
+                letter.innerText = text[i];
+
+                container.appendChild(letter);
+
+                setTimeout(function() {
+                    letter.style.opacity = '1';
+                }, 50);
+            }, i * 250);
+        })(i);
+    }
+
+    var gradioContainer = document.querySelector('.gradio-container');
+    gradioContainer.insertBefore(container, gradioContainer.firstChild);
+
+    return 'Animation created';
+}
+"""
+
+css = """
+#scrollable-content {
+    max-height: 200px;
+    overflow-y: auto !important;
+}
+#translate-bttn {
+  font-size: 30px;
+}
+
+/* Hiệu ứng khi lướt chuột qua */
+#translate-bttn:hover {
+  background-color: #00ff00; /* Nền xanh lá */
+  color: #000000;            /* Chữ đen */
+}
+
+/* Hiệu ứng khi nhấn chuột */
+#translate-bttn:active {
+  background-color: #008000; /* Nền xanh đậm */
+  color: #000000;            /* Chữ đen */
+}
+"""
+
+with gr.Blocks(js=js, css=css) as demo:
+    # Tạo ra các Block
+    with gr.Row(equal_height=False):
+        with gr.Column(scale=20):
+            with gr.Tab("Tiếng Việt") as in_vie_tab:
+                vie_in = gr.Textbox(label='', lines=3)
+            with gr.Tab("Tiếng Anh") as in_eng_tab:
+                eng_in = gr.Textbox(label='', lines=3)
+            with gr.Tab("Tiếng Lào") as in_lao_tab:
+                lao_in = gr.Textbox(label='', lines=3)
+            with gr.Tab("Tiếng Trung") as in_chi_tab:
+                chi_in = gr.Textbox(label='', lines=3)
+            with gr.Tab("Tiếng Khơ-me") as in_kho_tab:
+                kho_in = gr.Textbox(label='', lines=3)
+            in_selected = gr.Textbox(label='Ngôn ngữ', value='Tiếng Việt', visible=True)
+
+            with gr.Row():
+                models = gr.Radio(["Baseline", "Pretrained Model"],
+                                        label="Mô hình thực hiện",
+                                        value="Baseline",
+                                        interactive=True,
+                                        scale=10)
+                trans_bttn = gr.Button('Dịch', elem_id='translate-bttn')
+
+        with gr.Column(scale=20):
+            with gr.Tab("Tiếng Việt") as out_vie_tab:
+                vie_out = gr.Textbox(label='', lines=3, interactive=False)
+            with gr.Tab("Tiếng Anh") as out_eng_tab:
+                eng_out = gr.Textbox(label='', lines=3, interactive=False)
+            with gr.Tab("Tiếng Lào") as out_lao_tab:
+                lao_out = gr.Textbox(label='', lines=3, interactive=False)
+            with gr.Tab("Tiếng Trung") as out_chi_tab:
+                chi_out = gr.Textbox(label='', lines=3, interactive=False)
+            with gr.Tab("Tiếng Khơ-me") as out_kho_tab:
+                kho_out = gr.Textbox(label='', lines=3, interactive=False)
+            out_selected = gr.Textbox(label='Ngôn ngữ', value='Tiếng Việt', visible=True)
+
+            with gr.Row():
+                save_history = gr.Button('Lưu lịch sử')
+                clear_all_history = gr.Button('Xoá toàn bộ lịch sử')
+
+    with gr.Row():
+        default = gr.Button('Trạng thái mặc định')
+        clear_all = gr.Button('Xoá tất cả')
+
+    with gr.Row():
+        lang = gr.Textbox(label='Ngôn ngữ', visible=False)
+        sentence = gr.Textbox(label='Câu ví dụ', visible=False)
+        examples = gr.Examples(
+            examples=[
+                ['Tiếng Việt', 'Tôi là ai', 'Baseline'],
+                ['Tiếng Anh', 'The one who is called Kira is me', 'Baseline'],
+                ['Tiếng Lào', 'ຂ້ອຍ', 'Baseline'],
+                ['Tiếng Trung', '那头牛', 'Baseline'],
+                ['Tiếng Khơ-me', 'អ្នកណាយំឈឺចាប់នេះ?', 'Baseline'],
+                ['Tiếng Việt', '12 giờ đêm, khi những con bò lạc lối', 'Pretrained Model'],
+                ['Tiếng Anh', 'I love... your mother', 'Pretrained Model'],
+                ['Tiếng Lào', 'ດັ່ງທີ່ຄາດໄວ້, C-cup ຍັງບໍ່ພໍທີ່ຈະຊັກຈູງອ້າຍຂອງຂ້ອຍ', 'Pretrained Model'],
+                ['Tiếng Trung', '我哥哥被 E 罩杯的东西所吸引。', 'Pretrained Model'],
+                ['Tiếng Khơ-me', 'កុំមើល! វង្វេង!!', 'Pretrained Model'],
+            ],
+            inputs=[lang, sentence, models],
+            elem_id='scrollable-content',
+            label='Ví dụ',
+        )
+
+        src_sen = gr.Textbox(label='Câu nguồn', visible=False)
+        tgt_sen = gr.Textbox(label='Câu mục tiêu', visible=False)
+        histories = gr.Examples(
+            examples=history_list,
+            inputs=[src_sen, tgt_sen, models],
+            elem_id='scrollable-content',
+            label='Lịch sử',
+            cache_examples=False
+        )
+
+    # Xử lí phần sự kiện
+
+    ## Sự kiện nhấn nút dịch
+    gr.on(
+        triggers=[vie_in.submit, eng_in.submit, lao_in.submit, chi_in.submit, kho_in.submit, trans_bttn.click],
+        fn=translate,
+        inputs={vie_in, eng_in, lao_in, chi_in, kho_in, models, in_selected, out_selected},
+        outputs=[vie_out, eng_out, lao_out, chi_out, kho_out]
+    )
+
+    ## Sự kiện chuyển tab của phần Input
+    in_vie_tab.select(fn=change_input_tab,
+                   inputs={vie_in},
+                   outputs=[in_selected]
+                   )
+    in_eng_tab.select(fn=change_input_tab,
+                   inputs={eng_in},
+                   outputs=[in_selected]
+                   )
+    in_lao_tab.select(fn=change_input_tab,
+                   inputs={lao_in},
+                   outputs=[in_selected]
+                   )
+    in_chi_tab.select(fn=change_input_tab,
+                   inputs={chi_in},
+                   outputs=[in_selected]
+                   )
+    in_kho_tab.select(fn=change_input_tab,
+                   inputs={kho_in},
+                   outputs=[in_selected]
+                   )
+
+    ## Sự kiện ẩn chuyển tab của phần Output
+    out_vie_tab.select(fn=change_output_tab,
+                   inputs={vie_out},
+                   outputs=[out_selected]
+                   )
+    out_eng_tab.select(fn=change_output_tab,
+                   inputs={eng_out},
+                   outputs=[out_selected]
+                   )
+    out_lao_tab.select(fn=change_output_tab,
+                   inputs={lao_out},
+                   outputs=[out_selected]
+                   )
+    out_chi_tab.select(fn=change_output_tab,
+                   inputs={chi_out},
+                   outputs=[out_selected]
+                   )
+    out_kho_tab.select(fn=change_output_tab,
+                   inputs={kho_out},
+                   outputs=[out_selected]
+                   )
+
+    ## Sự kiện nút Mặc định
+    default.click(fn=default_func,
+                  inputs=None,
+                  outputs=[vie_in, eng_in, lao_in, chi_in, kho_in, models]
+                  )
+
+    ## Sự kiện xoá tất cả
+    clear_all.click(fn=clear_all_func,
+                    inputs=None,
+                    outputs=[vie_in, eng_in, lao_in, chi_in, kho_in, vie_out, eng_out, lao_out, chi_out, kho_out]
+                    )
+
+    ## Sự kiện click vô Ví dụ
+    sentence.change(fn=set_input,
+                    inputs={lang, sentence},
+                    outputs=[vie_in, eng_in, lao_in, chi_in, kho_in])
+
+    ## Sự kiện với Lịch sử
+    save_history.click(fn=add_history,
+                       inputs={vie_out, eng_out, lao_out, chi_out, kho_out, vie_in, eng_in, lao_in, chi_in, kho_in, models, in_selected, out_selected},
+                       outputs=histories.dataset)
+    clear_all_history.click(fn=clear_all_history_func,
+                            inputs=None,
+                            outputs=histories.dataset)
+
+demo.launch()

model/vi_en/Translation_en_vi_baseline/final-result/metric.txt

@@ -0,0 +1,2 @@
+bleu	ter	chrf	rouge1	rougeL
+12.053431657255456	92.91312867406923	22.964739702878658	0.31416478257591357	0.27341178815887657

model/vi_en/Translation_en_vi_baseline/log/model-en-vi-baseline.log

@@ -0,0 +1,273 @@
+8.2s 1 Collecting sacrebleu
+8.3s 2 Downloading sacrebleu-2.4.2-py3-none-any.whl.metadata (58 kB)
+8.3s 3 [?25l     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/58.0 kB ? eta -:--:--
+8.4s 4 [?25hCollecting portalocker (from sacrebleu)
+8.4s 5 Downloading portalocker-2.10.1-py3-none-any.whl.metadata (8.5 kB)
+8.5s 6 Requirement already satisfied: regex in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (2023.12.25)
+8.5s 7 Requirement already satisfied: tabulate>=0.8.9 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.9.0)
+8.5s 8 Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (1.26.4)
+8.5s 9 Requirement already satisfied: colorama in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.4.6)
+8.5s 10 Requirement already satisfied: lxml in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (5.2.2)
+8.5s 11 Downloading sacrebleu-2.4.2-py3-none-any.whl (106 kB)
+8.6s 12 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/106.7 kB ? eta -:--:--
+8.6s 13 [?25hDownloading portalocker-2.10.1-py3-none-any.whl (18 kB)
+19.1s 14 Installing collected packages: portalocker, sacrebleu
+19.4s 15 Successfully installed portalocker-2.10.1 sacrebleu-2.4.2
+20.7s 16 Collecting rouge-score
+20.8s 17 Downloading rouge_score-0.1.2.tar.gz (17 kB)
+21.8s 18 Preparing metadata (setup.py) ... [?25l- done
+21.8s 19 [?25hRequirement already satisfied: absl-py in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.4.0)
+21.8s 20 Requirement already satisfied: nltk in /opt/conda/lib/python3.10/site-packages (from rouge-score) (3.2.4)
+21.8s 21 Requirement already satisfied: numpy in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.26.4)
+21.8s 22 Requirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.16.0)
+21.8s 23 Building wheels for collected packages: rouge-score
+23.2s 24 Building wheel for rouge-score (setup.py) ... [?25l- \ done
+23.2s 25 [?25h  Created wheel for rouge-score: filename=rouge_score-0.1.2-py3-none-any.whl size=24934 sha256=b6796ff028fdaad83a8b8a942afa0a0f19e0e8749b8bc281e26289acbe530bd1
+23.2s 26 Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4
+23.2s 27 Successfully built rouge-score
+33.0s 28 Installing collected packages: rouge-score
+33.1s 29 Successfully installed rouge-score-0.1.2
+33.4s 30 Note: you may need to restart the kernel to use updated packages.
+40.1s 31 2024-08-06 04:22:50.535640: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+40.1s 32 2024-08-06 04:22:50.535769: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+40.1s 33 2024-08-06 04:22:50.644904: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+40.3s 34 2024-08-06 04:22:50.535640: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+40.3s 35 2024-08-06 04:22:50.535769: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+40.3s 36 2024-08-06 04:22:50.644904: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+56.3s 37 /opt/conda/lib/python3.10/site-packages/transformers/tokenization_utils_base.py:4016: UserWarning: `as_target_tokenizer` is deprecated and will be removed in v5 of Transformers. You can tokenize your labels by using the argument `text_target` of the regular `__call__` method (either in the same call as your input texts if you use the same keyword arguments, or in a separate call.
+56.3s 38 warnings.warn(
+84.1s 39 wandb: W&B API key is configured. Use `wandb login --relogin` to force relogin
+84.1s 40 wandb: WARNING If you're specifying your api key in code, ensure this code is not shared publicly.
+84.1s 41 wandb: WARNING Consider setting the WANDB_API_KEY environment variable, or running `wandb login` from the command line.
+84.1s 42 wandb: Appending key for api.wandb.ai to your netrc file: /root/.netrc
+84.3s 43 /opt/conda/lib/python3.10/site-packages/transformers/training_args.py:1494: FutureWarning: `evaluation_strategy` is deprecated and will be removed in version 4.46 of 🤗 Transformers. Use `eval_strategy` instead
+84.3s 44 warnings.warn(
+85.5s 45 wandb: WARNING The `run_name` is currently set to the same value as `TrainingArguments.output_dir`. If this was not intended, please specify a different run name by setting the `TrainingArguments.run_name` parameter.
+85.5s 46 wandb: Currently logged in as: lananhmtms1 (lananhmtms1-mtms1). Use `wandb login --relogin` to force relogin
+100.8s 47 wandb: wandb version 0.17.5 is available!  To upgrade, please run:
+100.8s 48 wandb:  $ pip install wandb --upgrade
+100.8s 49 wandb: Tracking run with wandb version 0.17.4
+100.8s 50 wandb: Run data is saved locally in /kaggle/working/wandb/run-20240806_042336-o4fgwvql
+100.8s 51 wandb: Run `wandb offline` to turn off syncing.
+100.8s 52 wandb: Syncing run ./results-en-vi-base
+100.8s 53 wandb: ⭐️ View project at https://wandb.ai/lananhmtms1-mtms1/huggingface
+100.8s 54 wandb: 🚀 View run at https://wandb.ai/lananhmtms1-mtms1/huggingface/runs/o4fgwvql
+103.6s 55 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+103.6s 56 warnings.warn('Was asked to gather along dimension 0, but all '
+287.4s 57 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+287.4s 58 warnings.warn('Was asked to gather along dimension 0, but all '
+444.3s 59 Translated: C nhng ti iu ti c iu ó vi iu iu i
+473.3s 60 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+473.3s 61 warnings.warn('Was asked to gather along dimension 0, but all '
+658.4s 62 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+658.4s 63 warnings.warn('Was asked to gather along dimension 0, but all '
+814.6s 64 Translated: Chng ta ng ta i iu i vi ó và iu  v  ó
+844.0s 65 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+844.0s 66 warnings.warn('Was asked to gather along dimension 0, but all '
+1029.3s 67 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1029.3s 68 warnings.warn('Was asked to gather along dimension 0, but all '
+1186.5s 69 Translated: ó có mt nhng tôi v v  nhng tôi  ó, ó
+1216.0s 70 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1216.0s 71 warnings.warn('Was asked to gather along dimension 0, but all '
+1401.6s 72 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1401.6s 73 warnings.warn('Was asked to gather along dimension 0, but all '
+1557.8s 74 Translated: Chng tôi là mt mt tng tng i c vi ó   ó i
+1587.1s 75 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1587.1s 76 warnings.warn('Was asked to gather along dimension 0, but all '
+1773.4s 77 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1773.4s 78 warnings.warn('Was asked to gather along dimension 0, but all '
+1930.5s 79 Translated: Chng tôi là mt mt tr  vi     ó   ti  có
+1959.9s 80 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1959.9s 81 warnings.warn('Was asked to gather along dimension 0, but all '
+2145.7s 82 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2145.7s 83 warnings.warn('Was asked to gather along dimension 0, but all '
+2302.7s 84 Translated: Chng tôi làm tôi c nhng và ca  ây, và ây là, v
+2332.7s 85 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2332.7s 86 warnings.warn('Was asked to gather along dimension 0, but all '
+2519.7s 87 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2519.7s 88 warnings.warn('Was asked to gather along dimension 0, but all '
+2676.9s 89 Translated: Chng ta là mt chng ta  tôi v v ó  ó, v nhng c
+2706.6s 90 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2706.6s 91 warnings.warn('Was asked to gather along dimension 0, but all '
+2893.3s 92 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2893.3s 93 warnings.warn('Was asked to gather along dimension 0, but all '
+3050.8s 94 Translated: Chng ta là nhng t cn th nhng v iu ó, iu
+3080.8s 95 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3080.8s 96 warnings.warn('Was asked to gather along dimension 0, but all '
+3266.8s 97 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3266.8s 98 warnings.warn('Was asked to gather along dimension 0, but all '
+3423.9s 99 Translated: Chng tôi là nhng ngi nht và ó là ông c  vi có nhng
+3453.8s 100 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3453.8s 101 warnings.warn('Was asked to gather along dimension 0, but all '
+3640.8s 102 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3640.8s 103 warnings.warn('Was asked to gather along dimension 0, but all '
+3798.1s 104 Translated: Chng tôi là mt nhng ngi nhng th và nó, v  nó cn
+3828.2s 105 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3828.2s 106 warnings.warn('Was asked to gather along dimension 0, but all '
+4014.8s 107 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4014.8s 108 warnings.warn('Was asked to gather along dimension 0, but all '
+4172.5s 109 Translated: Chng tôi to c c to v   nhng  ó, v th ca
+4202.4s 110 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4202.4s 111 warnings.warn('Was asked to gather along dimension 0, but all '
+4389.9s 112 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4389.9s 113 warnings.warn('Was asked to gather along dimension 0, but all '
+4547.7s 114 Translated: Chng ta là nhng ngi nh v v nó là  cnh th th thy
+4577.6s 115 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4577.6s 116 warnings.warn('Was asked to gather along dimension 0, but all '
+4764.2s 117 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4764.2s 118 warnings.warn('Was asked to gather along dimension 0, but all '
+4922.1s 119 Translated: chng tôi là nhng nhng th nghim và nh ó, v ó là ông
+4952.1s 120 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4952.1s 121 warnings.warn('Was asked to gather along dimension 0, but all '
+5139.7s 122 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5139.7s 123 warnings.warn('Was asked to gather along dimension 0, but all '
+5297.5s 124 Translated: Chng ta là nhng ngi nghiên t nht và ó, v vy cn c
+5327.6s 125 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5327.6s 126 warnings.warn('Was asked to gather along dimension 0, but all '
+5514.8s 127 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5514.8s 128 warnings.warn('Was asked to gather along dimension 0, but all '
+5672.5s 129 Translated: Chng ta là nhng nhng  và ti ó, nhng ng th s thc
+5702.5s 130 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5702.5s 131 warnings.warn('Was asked to gather along dimension 0, but all '
+5890.7s 132 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5890.7s 133 warnings.warn('Was asked to gather along dimension 0, but all '
+6049.1s 134 Translated: Chng ta là mt ngi trng v v nhng nhà th và iu  t
+6079.2s 135 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6079.2s 136 warnings.warn('Was asked to gather along dimension 0, but all '
+6267.4s 137 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6267.4s 138 warnings.warn('Was asked to gather along dimension 0, but all '
+6425.9s 139 Translated: Chng ta là nhng ngi hi ng và iu nh vy, vy, th
+6456.1s 140 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6456.1s 141 warnings.warn('Was asked to gather along dimension 0, but all '
+6644.1s 142 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6644.1s 143 warnings.warn('Was asked to gather along dimension 0, but all '
+6802.2s 144 Translated: Chng ta ng nhng ngi nh nhng ngi ng và ó là ông, v
+6832.3s 145 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6832.3s 146 warnings.warn('Was asked to gather along dimension 0, but all '
+7020.0s 147 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7020.0s 148 warnings.warn('Was asked to gather along dimension 0, but all '
+7177.8s 149 Translated: Chng tôi là nhng nhà nhà th và ây là ông tôi, vy, cô
+7208.4s 150 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7208.4s 151 warnings.warn('Was asked to gather along dimension 0, but all '
+7395.7s 152 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7395.7s 153 warnings.warn('Was asked to gather along dimension 0, but all '
+7553.1s 154 Translated: Chng ta có nhng ngôn ht vi và iu nh ó là ông tr
+7583.3s 155 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7583.3s 156 warnings.warn('Was asked to gather along dimension 0, but all '
+7771.0s 157 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7771.0s 158 warnings.warn('Was asked to gather along dimension 0, but all '
+7928.6s 159 Translated: Chng là nhà th nghim và cn ó là ông tr thành công s thch
+7958.8s 160 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7958.8s 161 warnings.warn('Was asked to gather along dimension 0, but all '
+8146.4s 162 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8146.4s 163 warnings.warn('Was asked to gather along dimension 0, but all '
+8304.5s 164 Translated: Chng ta là nhng ngi c c tt vt và v v nên ó, v
+8334.7s 165 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8334.7s 166 warnings.warn('Was asked to gather along dimension 0, but all '
+8522.5s 167 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8522.5s 168 warnings.warn('Was asked to gather along dimension 0, but all '
+8680.4s 169 Translated: Chng tôi là nhng ngh tht v v và ó là ông y, có th c
+8710.7s 170 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8710.7s 171 warnings.warn('Was asked to gather along dimension 0, but all '
+8898.0s 172 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8898.0s 173 warnings.warn('Was asked to gather along dimension 0, but all '
+9056.7s 174 Translated: Chng ta là nhng ng ngi t nht vy  ó, và y làm cn
+9086.9s 175 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9086.9s 176 warnings.warn('Was asked to gather along dimension 0, but all '
+9274.5s 177 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9274.5s 178 warnings.warn('Was asked to gather along dimension 0, but all '
+9432.1s 179 Translated: Chng ta là nhà khoa hc và ó là ông, vy nên cn hc s
+9462.7s 180 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9462.7s 181 warnings.warn('Was asked to gather along dimension 0, but all '
+9650.1s 182 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9650.1s 183 warnings.warn('Was asked to gather along dimension 0, but all '
+9808.1s 184 Translated: Chng ta là nhà nhà nhng ngi thut và nên cu c n c
+9838.8s 185 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9838.8s 186 warnings.warn('Was asked to gather along dimension 0, but all '
+10026.4s 187 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10026.4s 188 warnings.warn('Was asked to gather along dimension 0, but all '
+10184.8s 189 Translated: Chng ta là nhng th nghim tt và iu ó y, v vy c
+10215.5s 190 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10215.5s 191 warnings.warn('Was asked to gather along dimension 0, but all '
+10403.9s 192 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10403.9s 193 warnings.warn('Was asked to gather along dimension 0, but all '
+10561.5s 194 Translated: Chng tôi là nhà nhào tht v và ó là ông n, v v ông n
+10591.8s 195 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10591.8s 196 warnings.warn('Was asked to gather along dimension 0, but all '
+10780.0s 197 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10780.0s 198 warnings.warn('Was asked to gather along dimension 0, but all '
+10938.4s 199 Translated: Chng tôi là nhng tr nht và vy ông  ó, nên s s
+10970.0s 200 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10970.0s 201 warnings.warn('Was asked to gather along dimension 0, but all '
+11157.4s 202 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11157.4s 203 warnings.warn('Was asked to gather along dimension 0, but all '
+11315.3s 204 Translated: Chng ta là nhng ngi thc s tr ln và ó là, vy, ng
+11346.1s 205 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11346.1s 206 warnings.warn('Was asked to gather along dimension 0, but all '
+11533.6s 207 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11533.6s 208 warnings.warn('Was asked to gather along dimension 0, but all '
+11692.0s 209 Translated: Chng ta có nhng ngi nht và vy nó, v th nó s có
+11722.2s 210 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11722.2s 211 warnings.warn('Was asked to gather along dimension 0, but all '
+11909.8s 212 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11909.8s 213 warnings.warn('Was asked to gather along dimension 0, but all '
+12067.7s 214 Translated: Chng tôi là nhng nhà t nghin và v nên ông y, vy,
+12098.1s 215 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12098.1s 216 warnings.warn('Was asked to gather along dimension 0, but all '
+12286.0s 217 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12286.0s 218 warnings.warn('Was asked to gather along dimension 0, but all '
+12444.1s 219 Translated: Chng tôi là nhng nhà s tt v và nhng ngi ó là, vy
+12474.3s 220 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12474.3s 221 warnings.warn('Was asked to gather along dimension 0, but all '
+12661.9s 222 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12661.9s 223 warnings.warn('Was asked to gather along dimension 0, but all '
+12820.9s 224 Translated: Chng tôi là nhng nhà th và iu ó ó là c ngh thc hin
+12851.2s 225 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12851.2s 226 warnings.warn('Was asked to gather along dimension 0, but all '
+13039.5s 227 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13039.5s 228 warnings.warn('Was asked to gather along dimension 0, but all '
+13198.0s 229 Translated: Chng ta là nhng t tt v và cng th ni, th nên các
+13228.4s 230 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13228.4s 231 warnings.warn('Was asked to gather along dimension 0, but all '
+13416.2s 232 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13416.2s 233 warnings.warn('Was asked to gather along dimension 0, but all '
+13575.2s 234 Translated: Chng ta là nhng nhà bán tt và nó là ông y, thc s có th
+13605.3s 235 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13605.3s 236 warnings.warn('Was asked to gather along dimension 0, but all '
+13793.6s 237 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13793.6s 238 warnings.warn('Was asked to gather along dimension 0, but all '
+13951.0s 239 Translated: Chng ta là nhà tt nhng ngi th và n  nhiên, vy nh
+13981.7s 240 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13981.7s 241 warnings.warn('Was asked to gather along dimension 0, but all '
+14170.3s 242 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14170.3s 243 warnings.warn('Was asked to gather along dimension 0, but all '
+14327.6s 244 Translated: Chng ta là nhng ngi th nghim và ó là ông ta, vy, có th
+14357.7s 245 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14357.7s 246 warnings.warn('Was asked to gather along dimension 0, but all '
+14545.8s 247 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14545.8s 248 warnings.warn('Was asked to gather along dimension 0, but all '
+14703.5s 249 Translated: Chng tôi là nhng ngh thut và vy nó, vy, các phc
+14734.2s 250 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14734.2s 251 warnings.warn('Was asked to gather along dimension 0, but all '
+14922.5s 252 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14922.5s 253 warnings.warn('Was asked to gather along dimension 0, but all '
+15079.9s 254 Translated: Chng ta là nhng nhà khoa hc tt và cng nh vy, nhng
+15110.1s 255 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15110.1s 256 warnings.warn('Was asked to gather along dimension 0, but all '
+15297.5s 257 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15297.5s 258 warnings.warn('Was asked to gather along dimension 0, but all '
+15455.8s 259 Translated: Chng ta là nhng nhng nghip nht và iu này, v nên ô
+15486.4s 260 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15486.4s 261 warnings.warn('Was asked to gather along dimension 0, but all '
+15674.0s 262 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15674.0s 263 warnings.warn('Was asked to gather along dimension 0, but all '
+15725.1s 264 wandb: - 0.007 MB of 0.007 MB uploaded
+15725.1s 265 warn(
+15725.1s 266 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+15725.1s 267 [NbConvertApp] Converting notebook __notebook__.ipynb to notebook
+15725.6s 268 [NbConvertApp] Writing 199824 bytes to __notebook__.ipynb
+15727.2s 269 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["nbconvert.preprocessors.ExtractOutputPreprocessor"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+15727.2s 270 warn(
+15727.2s 271 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+15727.3s 272 [NbConvertApp] Converting notebook __notebook__.ipynb to html
+15728.3s 273 [NbConvertApp] Writing 456038 bytes to __results__.html

model/vi_en/Translation_en_vi_baseline/predict.py

@@ -0,0 +1,93 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+import pandas as pd
+
+SRC = 'en'
+TRG = 'vi'
+KIND = 'baseline'
+
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+#tải model:
+name_model = f"NguyenManhAI/translation-{SRC}-{TRG}-{KIND}"
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+if __name__ == "__main__":
+    sentence = {
+        'en': "I'm a doctor and so he is",
+        'vi': "Tôi là một bác sĩ và anh ấy cũng vậy."
+    }
+    
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(pred)
+
+    translation = {
+        'inputs':[],
+        'preds':[],
+        'labels':[]
+    }
+
+    for i in range(len(list_test[SRC])):
+        translation['inputs'].append(list_test[SRC][i])
+        translation['preds'].append(predict(model, list_test[SRC][i], tokenizer))
+        translation['labels'].append(list_test[TRG][i])
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(translation['preds'], [translation['labels']])
+    # Tính TER
+    ter = sacrebleu.corpus_ter(translation['preds'], [translation['labels']])
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(translation['preds'], [translation['labels']])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(translation['preds'], translation['labels']):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+    
+    metric_str = f"bleu\tter\tchrf\trouge1\trougeL\n{bleu.score}\t{ter.score}\t{chrf.score}\t{avg_rouge1}\t{avg_rougeL}"
+
+    f = open('final-result\metric.txt', 'w', encoding='utf-8')
+    f.write(metric_str)
+    f.close()
+
+    pd.DataFrame(translation).to_csv('final-result/translation.csv', index=False)
+
+    print("Lưu thành công")

model/vi_en/Translation_en_vi_baseline/result/log-step.csv

@@ -0,0 +1,167 @@
+Step	Training Loss	Validation Loss	Bleu	Ter	Rouge1	Rougel
+250	2.627700	1.644197	0.065444	102.810470	0.064245	0.062051
+500	1.628200	1.484654	0.288390	97.650880	0.133554	0.122853
+750	1.500700	1.417847	0.354922	95.965540	0.156740	0.143596
+1000	1.456000	1.378341	0.526384	94.247246	0.175121	0.161129
+1250	1.411100	1.349486	0.554477	93.729404	0.173738	0.161623
+1500	1.385600	1.325605	0.723026	93.122117	0.181949	0.168042
+1750	1.363100	1.303805	0.823366	92.420676	0.187867	0.174535
+2000	1.341900	1.287470	0.835699	91.884003	0.200039	0.184544
+2250	1.320500	1.271032	1.002635	91.031918	0.211180	0.193990
+2500	1.298800	1.258313	1.163212	90.937765	0.209801	0.192669
+2750	1.294300	1.243220	1.153392	90.203371	0.217644	0.199842
+3000	1.281100	1.228983	1.550688	89.370116	0.233425	0.211557
+3250	1.267900	1.220164	1.675214	89.068826	0.239106	0.216840
+3500	1.255400	1.205667	1.588106	88.645137	0.245444	0.223602
+3750	1.233600	1.196426	1.820132	88.325016	0.257691	0.232497
+4000	1.231600	1.183769	1.934878	87.920158	0.263296	0.238874
+4250	1.228300	1.172010	2.287752	87.265794	0.277792	0.250632
+4500	1.210700	1.161224	2.343722	86.922135	0.282720	0.255034
+4750	1.202400	1.153722	2.513547	86.456078	0.291601	0.261398
+5000	1.203500	1.140624	2.656975	85.942943	0.298051	0.268448
+5250	1.186000	1.131168	2.922282	85.716976	0.307243	0.275083
+5500	1.176700	1.121175	2.936947	85.213257	0.311061	0.278016
+5750	1.170600	1.112464	3.180533	85.076735	0.315466	0.282364
+6000	1.159600	1.105469	3.323173	84.888429	0.319794	0.286185
+6250	1.158800	1.097890	3.658102	83.979851	0.331756	0.296160
+6500	1.144200	1.090150	3.695026	83.880990	0.330546	0.296708
+6750	1.140400	1.085703	3.923268	83.509086	0.333630	0.298861
+7000	1.127600	1.077098	3.979903	83.579701	0.335560	0.300850
+7250	1.130600	1.067477	4.094874	83.137181	0.342696	0.306930
+7500	1.128700	1.063897	4.312253	82.755861	0.347492	0.310791
+7750	1.120800	1.057359	4.397489	82.581678	0.349256	0.312309
+8000	1.121200	1.052315	4.640964	82.657000	0.349732	0.313232
+8250	1.104300	1.046337	4.787352	82.332172	0.355654	0.317175
+8500	1.104100	1.041071	4.899065	82.181527	0.353432	0.317395
+8750	1.088300	1.034734	5.154683	81.668393	0.362565	0.323850
+9000	1.087000	1.028317	5.367251	81.343565	0.368787	0.329085
+9250	1.085700	1.026232	5.270144	81.640147	0.366218	0.327269
+9500	1.076600	1.020324	5.494634	81.254119	0.373639	0.332814
+9750	1.085400	1.014452	5.601409	81.207043	0.372757	0.333228
+10000	1.072300	1.008819	5.753288	81.004614	0.375827	0.336124
+10250	1.062700	1.003951	5.824650	80.910460	0.377952	0.337083
+10500	1.057900	0.997902	5.902168	80.821015	0.377506	0.337929
+10750	1.054100	0.994118	5.981207	80.453818	0.382981	0.342387
+11000	1.047600	0.990646	6.076846	80.279635	0.387679	0.346535
+11250	1.059900	0.984334	6.084267	80.223143	0.388361	0.347105
+11500	1.052700	0.981650	6.292561	80.016006	0.391179	0.350088
+11750	1.044200	0.976151	6.273452	80.133697	0.390527	0.349778
+12000	1.047100	0.973138	6.509788	79.870069	0.393277	0.352682
+12250	1.034100	0.970219	6.706573	79.592317	0.398129	0.355785
+12500	1.031600	0.966992	6.675276	79.375765	0.399087	0.356270
+12750	1.021400	0.961170	6.745674	79.187459	0.403253	0.360308
+13000	1.014900	0.959967	6.958013	79.210997	0.403968	0.360029
+13250	1.020200	0.956045	7.025415	79.107429	0.406925	0.361880
+13500	1.015900	0.953189	7.195114	79.145090	0.407031	0.362524
+13750	1.009300	0.950787	7.065260	78.923830	0.406814	0.362172
+14000	1.009800	0.943621	7.132338	78.905000	0.407837	0.364725
+14250	1.010100	0.942448	7.317758	78.890877	0.410191	0.367012
+14500	1.001300	0.938198	7.403996	78.490726	0.413053	0.369490
+14750	1.005600	0.935948	7.478498	78.443649	0.410890	0.367736
+15000	1.001800	0.931559	7.604042	78.401280	0.412341	0.368878
+15250	1.002300	0.927953	7.688558	78.325958	0.413995	0.371106
+15500	0.996300	0.926708	7.766576	78.147067	0.415794	0.372567
+15750	0.996100	0.923712	7.919211	78.245928	0.418260	0.373566
+16000	0.986600	0.920364	7.955050	78.109406	0.419570	0.375393
+16250	0.988200	0.917224	8.075858	77.582149	0.425575	0.380995
+16500	0.985500	0.914425	8.194338	77.859900	0.423397	0.379536
+16750	0.984800	0.913054	8.260551	77.596272	0.424273	0.381631
+17000	0.978500	0.910439	8.415732	77.219659	0.427690	0.383362
+17250	0.982300	0.908334	8.384414	77.384427	0.427202	0.382812
+17500	0.979200	0.904406	8.432118	77.271443	0.430293	0.385096
+17750	0.977500	0.902485	8.568459	77.144337	0.430542	0.386266
+18000	0.958900	0.901156	8.576069	77.436211	0.429110	0.384368
+18250	0.959500	0.898233	8.829191	77.083137	0.431090	0.386581
+18500	0.962600	0.896664	8.811914	76.927785	0.429840	0.387147
+18750	0.966600	0.894571	8.856628	77.252613	0.432888	0.388250
+19000	0.956200	0.891675	8.893505	77.285566	0.429843	0.385496
+19250	0.966900	0.889444	9.003679	77.214952	0.432689	0.388286
+19500	0.954900	0.887456	9.124519	76.993692	0.434827	0.389973
+19750	0.957900	0.886005	9.227721	76.734771	0.434806	0.389821
+20000	0.945100	0.884125	9.209296	76.786555	0.435456	0.390133
+20250	0.948200	0.881953	9.268570	76.744186	0.436718	0.391658
+20500	0.943500	0.881168	9.319876	76.715940	0.433865	0.391167
+20750	0.940500	0.878809	9.256061	76.570003	0.438009	0.393140
+21000	0.942500	0.877089	9.405705	76.871293	0.436517	0.392365
+21250	0.942500	0.875001	9.465343	76.522926	0.436819	0.393018
+21500	0.934500	0.873649	9.472936	76.494680	0.441052	0.396147
+21750	0.944900	0.872259	9.390609	76.395820	0.440584	0.395528
+22000	0.932200	0.869487	9.414966	76.645325	0.439741	0.394476
+22250	0.946700	0.868051	9.584151	76.329912	0.441111	0.395307
+22500	0.931000	0.866385	9.510247	76.381697	0.441110	0.395367
+22750	0.927700	0.864872	9.610698	76.320497	0.440212	0.396336
+23000	0.929700	0.863968	9.718435	76.311082	0.442863	0.397021
+23250	0.930100	0.862223	9.850952	76.052161	0.445375	0.400322
+23500	0.926200	0.860721	9.936575	75.915639	0.443956	0.399638
+23750	0.925400	0.858712	9.967990	76.042746	0.446046	0.401264
+24000	0.925900	0.857639	10.096671	75.877978	0.446397	0.402017
+24250	0.925300	0.856143	10.175190	76.014500	0.444481	0.401000
+24500	0.926700	0.854629	10.117966	75.736748	0.445366	0.401312
+24750	0.918200	0.854712	10.217045	75.896808	0.445830	0.401858
+25000	0.922200	0.851939	10.202913	75.830901	0.447015	0.401962
+25250	0.921500	0.851179	10.360602	75.732040	0.448666	0.403928
+25500	0.911100	0.849694	10.272043	75.868562	0.447008	0.402194
+25750	0.918800	0.848693	10.276174	75.647303	0.449982	0.404951
+26000	0.913600	0.847645	10.345470	75.807363	0.449043	0.404762
+26250	0.922100	0.846621	10.462485	75.487242	0.451472	0.407892
+26500	0.910000	0.845007	10.428833	75.562565	0.448986	0.405839
+26750	0.905300	0.844070	10.413072	75.534319	0.451794	0.407535
+27000	0.905000	0.843697	10.540369	75.487242	0.451758	0.407736
+27250	0.913500	0.843070	10.560556	75.327182	0.452559	0.408576
+27500	0.906700	0.840387	10.629209	75.444873	0.452537	0.408393
+27750	0.902600	0.839736	10.544614	75.284813	0.453074	0.409466
+28000	0.908100	0.838810	10.660547	75.284813	0.455547	0.411442
+28250	0.899500	0.837738	10.643213	75.383674	0.453473	0.410070
+28500	0.907000	0.837646	10.650279	75.388382	0.453337	0.409119
+28750	0.903600	0.836335	10.739483	75.313059	0.455209	0.411472
+29000	0.911700	0.835543	10.758739	75.247152	0.455721	0.411514
+29250	0.900100	0.834555	10.757329	75.185952	0.454988	0.410909
+29500	0.892500	0.833318	10.858246	75.091799	0.454998	0.411348
+29750	0.897000	0.832903	10.880794	75.077676	0.457176	0.412951
+30000	0.899000	0.831623	10.780874	75.265982	0.455298	0.410280
+30250	0.893700	0.831053	10.812154	75.063553	0.456747	0.413286
+30500	0.896100	0.830248	10.855250	75.143583	0.457235	0.412929
+30750	0.900600	0.829723	10.943592	75.025892	0.457687	0.413874
+31000	0.888200	0.828807	10.802195	75.148291	0.456015	0.411811
+31250	0.897900	0.827812	10.814122	74.875247	0.457076	0.413507
+31500	0.899100	0.827135	10.891732	74.856417	0.458300	0.414424
+31750	0.893000	0.826509	10.931727	74.969400	0.459407	0.415042
+32000	0.892800	0.825865	11.039062	74.856417	0.459129	0.415572
+32250	0.887300	0.825274	10.983367	75.035307	0.458136	0.414745
+32500	0.894700	0.824349	10.905021	75.035307	0.458091	0.414885
+32750	0.898100	0.823543	10.969078	74.922324	0.459772	0.415440
+33000	0.894500	0.822916	10.972579	74.959985	0.458621	0.415112
+33250	0.890500	0.822863	11.068211	74.865832	0.459588	0.415884
+33500	0.891200	0.821890	11.029868	74.828171	0.460945	0.417081
+33750	0.890800	0.821309	11.052303	74.809340	0.461285	0.416666
+34000	0.884700	0.821189	11.087364	74.682233	0.461172	0.417708
+34250	0.890400	0.820674	11.025102	74.781094	0.459885	0.416007
+34500	0.890200	0.820371	10.992115	74.861124	0.459446	0.415366
+34750	0.887500	0.819892	11.067214	74.696356	0.461642	0.417344
+35000	0.888500	0.819480	11.199964	74.583373	0.462673	0.418558
+35250	0.885500	0.818967	11.154836	74.701064	0.461443	0.418270
+35500	0.888200	0.818257	11.157596	74.705772	0.460943	0.417297
+35750	0.883100	0.817896	11.136658	74.635157	0.462180	0.418523
+36000	0.881900	0.817461	11.221197	74.649280	0.461815	0.417901
+36250	0.879200	0.817413	11.269516	74.385651	0.462369	0.419192
+36500	0.882700	0.817116	11.270089	74.583373	0.462395	0.418929
+36750	0.890500	0.816068	11.305302	74.418605	0.464837	0.420454
+37000	0.880300	0.816323	11.315746	74.357405	0.463152	0.419814
+37250	0.878200	0.816297	11.335717	74.324452	0.463925	0.420552
+37500	0.879500	0.816331	11.295830	74.508050	0.462985	0.419643
+37750	0.873700	0.815917	11.279245	74.442143	0.462336	0.419049
+38000	0.881500	0.815693	11.342169	74.475097	0.463437	0.419622
+38250	0.888600	0.815161	11.307758	74.451558	0.463186	0.420100
+38500	0.876200	0.815078	11.326251	74.385651	0.464323	0.420679
+38750	0.884300	0.814706	11.398963	74.343282	0.464915	0.421150
+39000	0.880300	0.814615	11.309243	74.329159	0.464053	0.420119
+39250	0.882700	0.814623	11.328777	74.451558	0.462804	0.419380
+39500	0.874900	0.814376	11.408448	74.291498	0.464396	0.421024
+39750	0.873700	0.814213	11.347434	74.366820	0.463777	0.420346
+40000	0.875700	0.814003	11.359607	74.357405	0.464388	0.420920
+40250	0.886000	0.813967	11.349502	74.282083	0.463924	0.420757
+40500	0.884300	0.813862	11.363786	74.418605	0.463203	0.419916
+40750	0.872600	0.813927	11.354252	74.371528	0.463216	0.419772
+41000	0.877700	0.813894	11.336277	74.395066	0.463134	0.419653
+41250	0.890900	0.813850	11.334885	74.418605	0.463340	0.419721
+41500	0.878800	0.813816	11.346388	74.399774	0.463420	0.419656

model/vi_en/Translation_en_vi_baseline/train.py

@@ -0,0 +1,183 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+
+SRC = 'en'
+TRG = 'vi'
+KIND = 'baseline'
+name_model = "google-t5/t5-small"
+
+# load dữ liệu
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+list_train = dict()
+list_train[SRC] = load_file(f"../data/train/{SRC}.{SRC}")
+list_train[TRG] = load_file(f"../data/train/{TRG}.{TRG}")
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+list_dev = dict()
+list_dev[SRC] = load_file(f"../data/dev/{SRC}-2012.{SRC}")
+list_dev[TRG] = load_file(f"../data/dev/{TRG}-2012.{TRG}")
+
+
+dataset_train = Dataset.from_dict({"source": list_train[SRC], "target": list_train})
+dataset_test = Dataset.from_dict({"source": list_test[SRC], "target": list_test[TRG]})
+dataset_dev = Dataset.from_dict({"source": list_dev[SRC], "target": list_dev[TRG]})
+
+#tải model:
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+#tạo các phương thức cần thiết:
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+def tokenize_function(examples):
+    inputs = [ex for ex in examples['source']]
+    targets = [ex for ex in examples['target']]
+    model_inputs = tokenizer(inputs, max_length=80, truncation=True, padding="max_length") #80 -> 128
+
+    # Sử dụng tokenizer để mã hóa câu đích
+    with tokenizer.as_target_tokenizer():
+        labels = tokenizer(text_target = targets, max_length=80, truncation=True, padding="max_length") # 80 -> 128
+
+    # Thêm nhãn vào kết quả mã hóa
+    model_inputs["labels"] = labels["input_ids"]
+    
+    return model_inputs
+
+def compute_metrics(eval_pred):
+    predictions, labels = eval_pred
+    
+    predictions = np.array(predictions)
+    predictions[predictions < 0] = 0
+    predictions = predictions.tolist()
+    
+    
+    labels = np.array(labels)
+    labels[labels < 0] = 0
+    labels = labels.tolist()
+    
+    decoded_preds = tokenizer.batch_decode(predictions, skip_special_tokens=True)
+    decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(decoded_preds, [decoded_labels])
+
+    # Tính TER
+    ter = sacrebleu.corpus_ter(decoded_preds, [decoded_labels])
+
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(decoded_preds, [decoded_labels])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(decoded_preds, decoded_labels):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+
+    return {
+        "bleu": bleu.score,
+        "ter": ter.score,
+        "chrf": chrf.score,
+        "rouge1": avg_rouge1,
+        "rougeL": avg_rougeL
+    }
+
+# chuẩn bị dữ liệu:
+tokenized_train = dataset_train.map(tokenize_function, batched=True)
+tokenized_test = dataset_test.map(tokenize_function, batched=True)
+
+data_collator = DataCollatorForSeq2Seq(tokenizer, model=model)
+
+# thiết lập callback
+
+from transformers import TrainerCallback
+
+class CustomStepCallback(TrainerCallback):
+    def __init__(self, steps_interval, action_method):
+        self.steps_interval = steps_interval
+        self.action_method = action_method
+
+    def on_step_end(self, args, state, control, **kwargs):
+        # Thực hiện hành động sau mỗi steps_interval bước
+        if state.global_step % self.steps_interval == 0:
+            self.action_method(state.global_step)
+
+# Định nghĩa phương thức hành động
+def custom_action(step):
+    sentence = {
+        'vi': "Chúng tôi là những bác sĩ xuất sắc và anh ấy cũng vậy, do đó ca phẫu thuật chắc chắn sẽ thành công.",
+        'en': "We are excellent doctors and so is he, so the surgery will definitely be successful."
+    }
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(f"Translated: {pred}")
+
+# Khởi tạo callback với số bước và phương thức hành động
+custom_callback = CustomStepCallback(steps_interval=1000, action_method=custom_action)
+
+# thiết lập huấn luyện
+training_args = Seq2SeqTrainingArguments(
+    output_dir=f"./results-{SRC}-{TRG}-base",
+    evaluation_strategy="steps",
+    eval_steps=250, #100 -> 250 x
+    learning_rate=3e-5, #3 -> 2
+    per_device_train_batch_size=16, #8 -> 16
+    per_device_eval_batch_size=16, #8 -> 16
+    weight_decay=0.01,
+    save_total_limit=3,
+    num_train_epochs=10, # 1 -> 10 x
+    predict_with_generate=True,
+    generation_max_length=50,
+    save_steps=500,
+    logging_dir="./logs",          # Thư mục để lưu logs
+    logging_steps=250,  
+    fp16 = True
+)
+
+trainer = Seq2SeqTrainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_train, #dev -> train x
+    eval_dataset=tokenized_test,
+    tokenizer=tokenizer,
+    data_collator=data_collator,
+    compute_metrics=compute_metrics,
+    callbacks=[custom_callback]
+)
+
+if __name__ == "__main__":
+    # huấn luyện và lưu lại mô hình
+    trainer.train()
+    torch.save(model.state_dict(), f"{SRC}-{TRG}-parameters-{KIND}.pth")

model/vi_en/Translation_en_vi_official/final-result/metric.txt

@@ -0,0 +1,2 @@
+bleu	ter	chrf	rouge1	rougeL
+34.68588567842991	57.73113235556083	51.795557929371924	0.7530914794583412	0.6796403900330101

model/vi_en/Translation_en_vi_official/log/model-en-vi-official.log

@@ -0,0 +1,484 @@
+15.7s 1 Collecting sacrebleu
+15.8s 2 Downloading sacrebleu-2.4.2-py3-none-any.whl.metadata (58 kB)
+15.8s 3 [?25l     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/58.0 kB ? eta -:--:--
+15.9s 4 [?25hCollecting portalocker (from sacrebleu)
+16.0s 5 Downloading portalocker-2.10.1-py3-none-any.whl.metadata (8.5 kB)
+16.0s 6 Requirement already satisfied: regex in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (2023.12.25)
+16.0s 7 Requirement already satisfied: tabulate>=0.8.9 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.9.0)
+16.0s 8 Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (1.26.4)
+16.0s 9 Requirement already satisfied: colorama in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.4.6)
+16.0s 10 Requirement already satisfied: lxml in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (5.2.2)
+16.1s 11 Downloading sacrebleu-2.4.2-py3-none-any.whl (106 kB)
+16.1s 12 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/106.7 kB ? eta -:--:--
+16.2s 13 [?25hDownloading portalocker-2.10.1-py3-none-any.whl (18 kB)
+28.1s 14 Installing collected packages: portalocker, sacrebleu
+28.3s 15 Successfully installed portalocker-2.10.1 sacrebleu-2.4.2
+29.6s 16 Collecting rouge-score
+29.7s 17 Downloading rouge_score-0.1.2.tar.gz (17 kB)
+31.0s 18 Preparing metadata (setup.py) ... [?25l- done
+31.0s 19 [?25hRequirement already satisfied: absl-py in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.4.0)
+31.0s 20 Requirement already satisfied: nltk in /opt/conda/lib/python3.10/site-packages (from rouge-score) (3.2.4)
+31.0s 21 Requirement already satisfied: numpy in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.26.4)
+31.0s 22 Requirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.16.0)
+31.1s 23 Building wheels for collected packages: rouge-score
+32.6s 24 Building wheel for rouge-score (setup.py) ... [?25l- \ done
+32.6s 25 [?25h  Created wheel for rouge-score: filename=rouge_score-0.1.2-py3-none-any.whl size=24934 sha256=625525a4adc177f5a70448e898d65cb1b0a93d0349fa83637d1b06e4bed5c489
+32.6s 26 Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4
+32.6s 27 Successfully built rouge-score
+42.6s 28 Installing collected packages: rouge-score
+42.8s 29 Successfully installed rouge-score-0.1.2
+43.2s 30 Note: you may need to restart the kernel to use updated packages.
+56.9s 31 2024-08-04 08:17:09.629217: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+56.9s 32 2024-08-04 08:17:09.629369: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+56.9s 33 2024-08-04 08:17:09.903206: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+57.1s 34 2024-08-04 08:17:09.629217: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+57.1s 35 2024-08-04 08:17:09.629369: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+57.1s 36 2024-08-04 08:17:09.903206: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+80.4s 37 /opt/conda/lib/python3.10/site-packages/torch/_utils.py:831: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
+80.4s 38 return self.fget.__get__(instance, owner)()
+83.6s 39 /opt/conda/lib/python3.10/site-packages/transformers/models/marian/tokenization_marian.py:175: UserWarning: Recommended: pip install sacremoses.
+83.6s 40 warnings.warn("Recommended: pip install sacremoses.")
+85.3s 41 /opt/conda/lib/python3.10/site-packages/transformers/tokenization_utils_base.py:4016: UserWarning: `as_target_tokenizer` is deprecated and will be removed in v5 of Transformers. You can tokenize your labels by using the argument `text_target` of the regular `__call__` method (either in the same call as your input texts if you use the same keyword arguments, or in a separate call.
+85.3s 42 warnings.warn(
+144.2s 43 wandb: W&B API key is configured. Use `wandb login --relogin` to force relogin
+144.2s 44 wandb: WARNING If you're specifying your api key in code, ensure this code is not shared publicly.
+144.2s 45 wandb: WARNING Consider setting the WANDB_API_KEY environment variable, or running `wandb login` from the command line.
+144.2s 46 wandb: Appending key for api.wandb.ai to your netrc file: /root/.netrc
+144.4s 47 /opt/conda/lib/python3.10/site-packages/transformers/training_args.py:1494: FutureWarning: `evaluation_strategy` is deprecated and will be removed in version 4.46 of 🤗 Transformers. Use `eval_strategy` instead
+144.4s 48 warnings.warn(
+145.9s 49 wandb: WARNING The `run_name` is currently set to the same value as `TrainingArguments.output_dir`. If this was not intended, please specify a different run name by setting the `TrainingArguments.run_name` parameter.
+145.9s 50 wandb: Currently logged in as: phuctranmtms1 (phuctranmtms1-mtms1). Use `wandb login --relogin` to force relogin
+161.2s 51 wandb: wandb version 0.17.5 is available!  To upgrade, please run:
+161.2s 52 wandb:  $ pip install wandb --upgrade
+161.3s 53 wandb: Tracking run with wandb version 0.17.4
+161.3s 54 wandb: Run data is saved locally in /kaggle/working/wandb/run-20240804_081839-fzh1co5a
+161.3s 55 wandb: Run `wandb offline` to turn off syncing.
+161.3s 56 wandb: Syncing run ./results-en-vi
+161.3s 57 wandb: ⭐️ View project at https://wandb.ai/phuctranmtms1-mtms1/huggingface
+161.3s 58 wandb: 🚀 View run at https://wandb.ai/phuctranmtms1-mtms1/huggingface/runs/fzh1co5a
+164.3s 59 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+164.3s 60 warnings.warn('Was asked to gather along dimension 0, but all '
+549.5s 61 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+549.5s 62 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+551.6s 63 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+551.6s 64 warnings.warn('Was asked to gather along dimension 0, but all '
+850.9s 65 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy ca phẫu thuật chắc chắn sẽ thành công.
+942.2s 66 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+942.2s 67 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+944.0s 68 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+944.0s 69 warnings.warn('Was asked to gather along dimension 0, but all '
+1334.2s 70 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+1334.2s 71 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+1336.1s 72 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1336.1s 73 warnings.warn('Was asked to gather along dimension 0, but all '
+1636.3s 74 Translated: Cả hai chúng tôi đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+1729.1s 75 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+1729.1s 76 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+1731.0s 77 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1731.0s 78 warnings.warn('Was asked to gather along dimension 0, but all '
+2121.3s 79 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+2121.3s 80 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+2123.2s 81 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2123.2s 82 warnings.warn('Was asked to gather along dimension 0, but all '
+2423.8s 83 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+2515.7s 84 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+2515.7s 85 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+2517.7s 86 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2517.7s 87 warnings.warn('Was asked to gather along dimension 0, but all '
+2909.6s 88 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+2909.6s 89 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+2911.6s 90 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2911.6s 91 warnings.warn('Was asked to gather along dimension 0, but all '
+3211.5s 92 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+3303.4s 93 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+3303.4s 94 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+3305.4s 95 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3305.4s 96 warnings.warn('Was asked to gather along dimension 0, but all '
+3696.6s 97 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+3696.6s 98 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+3698.6s 99 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3698.6s 100 warnings.warn('Was asked to gather along dimension 0, but all '
+3998.2s 101 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+4089.6s 102 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+4089.6s 103 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+4091.5s 104 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4091.5s 105 warnings.warn('Was asked to gather along dimension 0, but all '
+4483.7s 106 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+4483.7s 107 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+4485.6s 108 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4485.6s 109 warnings.warn('Was asked to gather along dimension 0, but all '
+4786.3s 110 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+4879.5s 111 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+4879.5s 112 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+4881.4s 113 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4881.4s 114 warnings.warn('Was asked to gather along dimension 0, but all '
+5271.5s 115 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+5271.5s 116 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+5273.4s 117 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5273.4s 118 warnings.warn('Was asked to gather along dimension 0, but all '
+5573.1s 119 Translated: Chúng ta đều là những bác sỹ xuất sắc, và ông ấy cũng vậy. Vì vậy, cuộc phẫu thuật chắc chắn sẽ thành công.
+5665.4s 120 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+5665.4s 121 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+5667.3s 122 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5667.3s 123 warnings.warn('Was asked to gather along dimension 0, but all '
+6058.3s 124 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+6058.3s 125 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+6060.2s 126 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6060.2s 127 warnings.warn('Was asked to gather along dimension 0, but all '
+6360.9s 128 Translated: Chúng ta đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vậy là ca phẫu thuật chắc chắn sẽ thành công.
+6453.4s 129 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+6453.4s 130 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+6455.4s 131 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6455.4s 132 warnings.warn('Was asked to gather along dimension 0, but all '
+6845.9s 133 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+6845.9s 134 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+6847.8s 135 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6847.8s 136 warnings.warn('Was asked to gather along dimension 0, but all '
+7148.5s 137 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy nên cuộc phẫu thuật chắc chắn sẽ thành công.
+7239.8s 138 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+7239.8s 139 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+7241.8s 140 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7241.8s 141 warnings.warn('Was asked to gather along dimension 0, but all '
+7635.5s 142 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+7635.5s 143 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+7637.5s 144 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7637.5s 145 warnings.warn('Was asked to gather along dimension 0, but all '
+7938.3s 146 Translated: Cả hai chúng tôi đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Vì vậy, cuộc phẫu thuật chắc chắn sẽ thành công.
+8029.1s 147 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+8029.1s 148 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+8031.1s 149 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8031.1s 150 warnings.warn('Was asked to gather along dimension 0, but all '
+8424.2s 151 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+8424.2s 152 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+8426.1s 153 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8426.1s 154 warnings.warn('Was asked to gather along dimension 0, but all '
+8725.2s 155 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy nên cuộc phẫu thuật chắc chắn sẽ thành công.
+8816.1s 156 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+8816.1s 157 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+8818.0s 158 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8818.0s 159 warnings.warn('Was asked to gather along dimension 0, but all '
+9209.4s 160 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+9209.4s 161 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+9211.3s 162 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9211.3s 163 warnings.warn('Was asked to gather along dimension 0, but all '
+9511.8s 164 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ta cũng thế. Ca phẫu thuật chắc chắn sẽ thành công.
+9604.4s 165 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+9604.4s 166 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+9606.3s 167 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9606.3s 168 warnings.warn('Was asked to gather along dimension 0, but all '
+9999.4s 169 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+9999.4s 170 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+10001.2s 171 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10001.2s 172 warnings.warn('Was asked to gather along dimension 0, but all '
+10301.8s 173 Translated: Chúng ta đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vậy nên cuộc phẫu thuật chắc chắn sẽ thành công.
+10394.9s 174 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+10394.9s 175 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+10396.9s 176 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10396.9s 177 warnings.warn('Was asked to gather along dimension 0, but all '
+10790.5s 178 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+10790.5s 179 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+10792.5s 180 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10792.5s 181 warnings.warn('Was asked to gather along dimension 0, but all '
+11092.2s 182 Translated: Chúng ta đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+11184.5s 183 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+11184.5s 184 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+11186.5s 185 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11186.5s 186 warnings.warn('Was asked to gather along dimension 0, but all '
+11578.1s 187 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+11578.1s 188 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+11580.1s 189 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11580.1s 190 warnings.warn('Was asked to gather along dimension 0, but all '
+11880.1s 191 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy nên cuộc phẫu thuật chắc chắn sẽ thành công.
+11973.3s 192 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+11973.3s 193 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+11975.3s 194 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11975.3s 195 warnings.warn('Was asked to gather along dimension 0, but all '
+12367.8s 196 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+12367.8s 197 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+12369.8s 198 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12369.8s 199 warnings.warn('Was asked to gather along dimension 0, but all '
+12670.1s 200 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Vì vậy ca phẫu thuật chắc chắn sẽ thành công.
+12762.6s 201 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+12762.6s 202 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+12764.6s 203 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12764.6s 204 warnings.warn('Was asked to gather along dimension 0, but all '
+13160.1s 205 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+13160.1s 206 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+13162.1s 207 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13162.1s 208 warnings.warn('Was asked to gather along dimension 0, but all '
+13463.5s 209 Translated: Cả hai chúng tôi đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+13556.7s 210 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+13556.7s 211 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+13558.6s 212 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13558.6s 213 warnings.warn('Was asked to gather along dimension 0, but all '
+13951.2s 214 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+13951.2s 215 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+13953.1s 216 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13953.1s 217 warnings.warn('Was asked to gather along dimension 0, but all '
+14253.3s 218 Translated: Chúng ta đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Vì vậy, ca phẫu thuật chắc chắn sẽ thành công.
+14345.8s 219 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+14345.8s 220 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+14347.7s 221 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14347.7s 222 warnings.warn('Was asked to gather along dimension 0, but all '
+14740.1s 223 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+14740.1s 224 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+14742.1s 225 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14742.1s 226 warnings.warn('Was asked to gather along dimension 0, but all '
+15042.5s 227 Translated: Cả hai chúng tôi đều là những bác sĩ tuyệt vời, và anh ta cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+15134.5s 228 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+15134.5s 229 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+15136.4s 230 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15136.4s 231 warnings.warn('Was asked to gather along dimension 0, but all '
+15527.7s 232 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+15527.7s 233 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+15529.6s 234 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15529.6s 235 warnings.warn('Was asked to gather along dimension 0, but all '
+15831.6s 236 Translated: Chúng ta đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vì vậy, ca phẫu thuật chắc chắn sẽ thành công.
+15925.4s 237 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+15925.4s 238 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+15927.3s 239 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15927.3s 240 warnings.warn('Was asked to gather along dimension 0, but all '
+16317.7s 241 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+16317.7s 242 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+16319.6s 243 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16319.6s 244 warnings.warn('Was asked to gather along dimension 0, but all '
+16621.1s 245 Translated: Cả hai chúng tôi đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+16713.9s 246 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+16713.9s 247 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+16715.9s 248 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16715.9s 249 warnings.warn('Was asked to gather along dimension 0, but all '
+17109.4s 250 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+17109.4s 251 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+17111.4s 252 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+17111.4s 253 warnings.warn('Was asked to gather along dimension 0, but all '
+17412.3s 254 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Vậy nên cuộc phẫu thuật chắc chắn sẽ thành công.
+17504.6s 255 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+17504.6s 256 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+17506.5s 257 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+17506.5s 258 warnings.warn('Was asked to gather along dimension 0, but all '
+17900.8s 259 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+17900.8s 260 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+17902.8s 261 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+17902.8s 262 warnings.warn('Was asked to gather along dimension 0, but all '
+18203.2s 263 Translated: Chúng tôi đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vì vậy, cuộc phẫu thuật chắc chắn sẽ thành công.
+18296.4s 264 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+18296.4s 265 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+18298.3s 266 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+18298.3s 267 warnings.warn('Was asked to gather along dimension 0, but all '
+18692.7s 268 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+18692.7s 269 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+18694.7s 270 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+18694.7s 271 warnings.warn('Was asked to gather along dimension 0, but all '
+18996.1s 272 Translated: Cả hai chúng tôi đều là những bác sĩ tuyệt vời, và anh ấy cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+19089.3s 273 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+19089.3s 274 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+19091.4s 275 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+19091.4s 276 warnings.warn('Was asked to gather along dimension 0, but all '
+19485.5s 277 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+19485.5s 278 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+19487.4s 279 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+19487.4s 280 warnings.warn('Was asked to gather along dimension 0, but all '
+19788.3s 281 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vì vậy ca phẫu thuật chắc chắn sẽ thành công.
+19880.4s 282 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+19880.4s 283 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+19882.4s 284 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+19882.4s 285 warnings.warn('Was asked to gather along dimension 0, but all '
+20276.3s 286 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+20276.3s 287 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+20278.3s 288 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+20278.3s 289 warnings.warn('Was asked to gather along dimension 0, but all '
+20578.1s 290 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+20670.4s 291 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+20670.4s 292 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+20672.3s 293 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+20672.3s 294 warnings.warn('Was asked to gather along dimension 0, but all '
+21066.0s 295 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+21066.0s 296 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+21068.0s 297 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+21068.0s 298 warnings.warn('Was asked to gather along dimension 0, but all '
+21369.2s 299 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+21463.2s 300 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+21463.2s 301 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+21465.1s 302 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+21465.1s 303 warnings.warn('Was asked to gather along dimension 0, but all '
+21859.4s 304 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+21859.4s 305 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+21861.4s 306 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+21861.4s 307 warnings.warn('Was asked to gather along dimension 0, but all '
+22162.1s 308 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+22255.3s 309 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+22255.3s 310 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+22257.3s 311 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+22257.3s 312 warnings.warn('Was asked to gather along dimension 0, but all '
+22649.9s 313 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+22649.9s 314 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+22651.8s 315 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+22651.8s 316 warnings.warn('Was asked to gather along dimension 0, but all '
+22952.7s 317 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+23046.6s 318 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+23046.6s 319 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+23048.5s 320 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+23048.5s 321 warnings.warn('Was asked to gather along dimension 0, but all '
+23440.4s 322 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+23440.4s 323 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+23442.4s 324 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+23442.4s 325 warnings.warn('Was asked to gather along dimension 0, but all '
+23742.7s 326 Translated: Cả hai chúng tôi đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+23835.5s 327 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+23835.5s 328 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+23837.4s 329 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+23837.4s 330 warnings.warn('Was asked to gather along dimension 0, but all '
+24230.6s 331 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+24230.6s 332 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+24232.6s 333 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+24232.6s 334 warnings.warn('Was asked to gather along dimension 0, but all '
+24533.9s 335 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+24627.1s 336 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+24627.1s 337 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+24629.0s 338 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+24629.0s 339 warnings.warn('Was asked to gather along dimension 0, but all '
+25022.9s 340 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+25022.9s 341 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+25024.8s 342 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+25024.8s 343 warnings.warn('Was asked to gather along dimension 0, but all '
+25325.6s 344 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vì vậy, cuộc phẫu thuật chắc chắn sẽ thành công.
+25418.9s 345 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+25418.9s 346 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+25420.8s 347 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+25420.8s 348 warnings.warn('Was asked to gather along dimension 0, but all '
+25815.2s 349 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+25815.2s 350 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+25817.5s 351 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+25817.5s 352 warnings.warn('Was asked to gather along dimension 0, but all '
+26117.5s 353 Translated: Chúng tôi đều là những bác sĩ xuất sắc, anh ấy cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+26209.8s 354 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+26209.8s 355 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+26212.1s 356 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+26212.1s 357 warnings.warn('Was asked to gather along dimension 0, but all '
+26607.2s 358 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+26607.2s 359 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+26609.2s 360 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+26609.2s 361 warnings.warn('Was asked to gather along dimension 0, but all '
+26910.4s 362 Translated: Chúng tôi đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+27003.8s 363 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+27003.8s 364 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+27005.7s 365 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+27005.7s 366 warnings.warn('Was asked to gather along dimension 0, but all '
+27397.7s 367 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+27397.7s 368 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+27399.6s 369 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+27399.6s 370 warnings.warn('Was asked to gather along dimension 0, but all '
+27700.6s 371 Translated: Chúng tôi đều là những bác sĩ tuyệt vời, và anh ấy cũng vậy. Vì vậy, cuộc phẫu thuật chắc chắn sẽ thành công.
+27793.0s 372 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+27793.0s 373 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+27794.9s 374 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+27794.9s 375 warnings.warn('Was asked to gather along dimension 0, but all '
+28188.7s 376 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+28188.7s 377 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+28190.7s 378 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+28190.7s 379 warnings.warn('Was asked to gather along dimension 0, but all '
+28491.6s 380 Translated: Chúng tôi đều là những bác sĩ xuất sắc, và anh ấy cũng vậy. Ca phẫu thuật chắc chắn sẽ thành công.
+28584.5s 381 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+28584.5s 382 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+28586.4s 383 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+28586.4s 384 warnings.warn('Was asked to gather along dimension 0, but all '
+28980.7s 385 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+28980.7s 386 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+28982.6s 387 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+28982.6s 388 warnings.warn('Was asked to gather along dimension 0, but all '
+29283.8s 389 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ta cũng vậy. Vì vậy, cuộc phẫu thuật chắc chắn sẽ thành công.
+29377.9s 390 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+29377.9s 391 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+29379.9s 392 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+29379.9s 393 warnings.warn('Was asked to gather along dimension 0, but all '
+29774.1s 394 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+29774.1s 395 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+29776.0s 396 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+29776.0s 397 warnings.warn('Was asked to gather along dimension 0, but all '
+30076.9s 398 Translated: Chúng ta đều là những bác sĩ xuất sắc, và anh ta cũng vậy. Vì vậy, ca phẫu thuật chắc chắn sẽ thành công.
+30170.3s 399 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+30170.3s 400 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+30172.2s 401 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+30172.2s 402 warnings.warn('Was asked to gather along dimension 0, but all '
+30565.8s 403 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+30565.8s 404 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+30567.8s 405 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+30567.8s 406 warnings.warn('Was asked to gather along dimension 0, but all '
+30868.4s 407 Translated: Cả hai chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+30961.2s 408 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+30961.2s 409 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+30963.1s 410 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+30963.1s 411 warnings.warn('Was asked to gather along dimension 0, but all '
+31356.3s 412 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+31356.3s 413 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+31358.2s 414 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+31358.2s 415 warnings.warn('Was asked to gather along dimension 0, but all '
+31658.7s 416 Translated: Chúng tôi đều là những bác sĩ giỏi, và anh ấy cũng vậy. Vì vậy, ca phẫu thuật chắc chắn sẽ thành công.
+31751.3s 417 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+31751.3s 418 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+31753.3s 419 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+31753.3s 420 warnings.warn('Was asked to gather along dimension 0, but all '
+32147.3s 421 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+32147.3s 422 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+32149.2s 423 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+32149.2s 424 warnings.warn('Was asked to gather along dimension 0, but all '
+32449.6s 425 Translated: Chúng ta đều là những bác sĩ giỏi, và anh ta cũng vậy. Vậy nên ca phẫu thuật chắc chắn sẽ thành công.
+32542.0s 426 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+32542.0s 427 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+32544.0s 428 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+32544.0s 429 warnings.warn('Was asked to gather along dimension 0, but all '
+32937.5s 430 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+32937.5s 431 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+32939.4s 432 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+32939.4s 433 warnings.warn('Was asked to gather along dimension 0, but all '
+33009.9s 434 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+33009.9s 435 Non-default generation parameters: {'max_length': 512, 'num_beams': 4, 'bad_words_ids': [[53684]], 'forced_eos_token_id': 0}
+33016.3s 436 Traceback (most recent call last):
+33016.3s 437 File "<string>", line 1, in <module>
+33016.3s 438 File "/opt/conda/lib/python3.10/site-packages/papermill/execute.py", line 134, in execute_notebook
+33016.3s 439 raise_for_execution_errors(nb, output_path)
+33016.3s 440 File "/opt/conda/lib/python3.10/site-packages/papermill/execute.py", line 241, in raise_for_execution_errors
+33016.3s 441 raise error
+33016.3s 442 papermill.exceptions.PapermillExecutionError:
+33016.3s 443 ---------------------------------------------------------------------------
+33016.3s 444 Exception encountered at "In [14]":
+33016.3s 445 ---------------------------------------------------------------------------
+33016.3s 446 FileNotFoundError                         Traceback (most recent call last)
+33016.3s 447 Cell In[14], line 1
+33016.3s 448 ----> 1 state_dict = torch.load('/kaggle/working/T5-en-vi-parameters.pth')
+33016.3s 449       3 # Cập nhật trọng số vào mô hình
+33016.3s 450       4 model_T5_en_vi.load_state_dict(state_dict)
+33016.3s 451 
+33016.3s 452 File /opt/conda/lib/python3.10/site-packages/torch/serialization.py:986, in load(f, map_location, pickle_module, weights_only, mmap, **pickle_load_args)
+33016.3s 453     983 if 'encoding' not in pickle_load_args.keys():
+33016.3s 454     984     pickle_load_args['encoding'] = 'utf-8'
+33016.3s 455 --> 986 with _open_file_like(f, 'rb') as opened_file:
+33016.3s 456     987     if _is_zipfile(opened_file):
+33016.3s 457     988         # The zipfile reader is going to advance the current file position.
+33016.3s 458     989         # If we want to actually tail call to torch.jit.load, we need to
+33016.3s 459     990         # reset back to the original position.
+33016.3s 460     991         orig_position = opened_file.tell()
+33016.3s 461 
+33016.3s 462 File /opt/conda/lib/python3.10/site-packages/torch/serialization.py:435, in _open_file_like(name_or_buffer, mode)
+33016.3s 463     433 def _open_file_like(name_or_buffer, mode):
+33016.3s 464     434     if _is_path(name_or_buffer):
+33016.3s 465 --> 435         return _open_file(name_or_buffer, mode)
+33016.3s 466     436     else:
+33016.3s 467     437         if 'w' in mode:
+33016.3s 468 
+33016.3s 469 File /opt/conda/lib/python3.10/site-packages/torch/serialization.py:416, in _open_file.__init__(self, name, mode)
+33016.3s 470     415 def __init__(self, name, mode):
+33016.3s 471 --> 416     super().__init__(open(name, mode))
+33016.3s 472 
+33016.3s 473 FileNotFoundError: [Errno 2] No such file or directory: '/kaggle/working/T5-en-vi-parameters.pth'
+33016.3s 474 
+33018.6s 475 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["remove_papermill_header.RemovePapermillHeader"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+33018.6s 476 warn(
+33018.7s 477 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+33018.7s 478 [NbConvertApp] Converting notebook __notebook__.ipynb to notebook
+33019.2s 479 [NbConvertApp] Writing 254414 bytes to __notebook__.ipynb
+33020.9s 480 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["nbconvert.preprocessors.ExtractOutputPreprocessor"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+33020.9s 481 warn(
+33020.9s 482 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+33021.0s 483 [NbConvertApp] Converting notebook __notebook__.ipynb to html
+33022.1s 484 [NbConvertApp] Writing 511436 bytes to __results__.html

model/vi_en/Translation_en_vi_official/predict.py

@@ -0,0 +1,93 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+import pandas as pd
+
+SRC = 'en'
+TRG = 'vi'
+KIND = 'official'
+
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+#tải model:
+name_model = f"NguyenManhAI/translation-{SRC}-{TRG}-{KIND}"
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+if __name__ == "__main__":
+    sentence = {
+        'en': "I'm a doctor and so he is",
+        'vi': "Tôi là một bác sĩ và anh ấy cũng vậy."
+    }
+    
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(pred)
+
+    translation = {
+        'inputs':[],
+        'preds':[],
+        'labels':[]
+    }
+
+    for i in range(len(list_test[SRC])):
+        translation['inputs'].append(list_test[SRC][i])
+        translation['preds'].append(predict(model, list_test[SRC][i], tokenizer))
+        translation['labels'].append(list_test[TRG][i])
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(translation['preds'], [translation['labels']])
+    # Tính TER
+    ter = sacrebleu.corpus_ter(translation['preds'], [translation['labels']])
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(translation['preds'], [translation['labels']])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(translation['preds'], translation['labels']):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+    
+    metric_str = f"bleu\tter\tchrf\trouge1\trougeL\n{bleu.score}\t{ter.score}\t{chrf.score}\t{avg_rouge1}\t{avg_rougeL}"
+
+    f = open('final-result\metric.txt', 'w', encoding='utf-8')
+    f.write(metric_str)
+    f.close()
+
+    pd.DataFrame(translation).to_csv('final-result/translation.csv', index=False)
+
+    print("Lưu thành công")

model/vi_en/Translation_en_vi_official/result/log-step.csv

@@ -0,0 +1,167 @@
+Step	Training Loss	Validation Loss	Bleu	Ter	Rouge1	Rougel
+250	0.419000	0.290262	34.528427	52.313225	0.752548	0.678462
+500	0.326400	0.286587	34.800746	52.096158	0.756820	0.681762
+750	0.315700	0.284458	34.896314	52.173913	0.754367	0.679986
+1000	0.318800	0.284286	34.799413	51.976285	0.752091	0.679039
+1250	0.315800	0.283449	35.001685	52.047560	0.755533	0.681133
+1500	0.316300	0.283837	34.437615	52.533532	0.751946	0.677418
+1750	0.311000	0.283472	35.142027	51.914728	0.755751	0.682285
+2000	0.313600	0.282538	34.885734	52.180393	0.754830	0.680829
+2250	0.309500	0.282026	35.073395	52.209551	0.753928	0.680840
+2500	0.306900	0.282315	35.002420	51.730059	0.756302	0.682668
+2750	0.310000	0.281654	35.056495	51.846692	0.754232	0.680863
+3000	0.311000	0.281669	34.930082	52.083198	0.753351	0.679717
+3250	0.311700	0.281589	34.891642	52.047560	0.755611	0.680720
+3500	0.303900	0.281686	34.820813	52.381261	0.755245	0.680496
+3750	0.305800	0.282299	34.703310	52.335904	0.754166	0.679429
+4000	0.305000	0.282130	34.979070	51.930927	0.754864	0.681688
+4250	0.303900	0.282648	34.665182	52.216031	0.755878	0.681373
+4500	0.284700	0.282799	34.779346	52.177153	0.755151	0.681280
+4750	0.283500	0.282231	34.575636	52.329424	0.755291	0.681444
+5000	0.287800	0.282950	35.066532	52.066999	0.754510	0.681349
+5250	0.282400	0.282641	34.738046	52.449297	0.755078	0.680114
+5500	0.281700	0.282985	35.103704	52.115596	0.755229	0.681769
+5750	0.285000	0.282623	34.993686	52.284067	0.753834	0.680014
+6000	0.284600	0.282732	34.879166	52.342383	0.753729	0.678728
+6250	0.288300	0.282963	34.786185	52.180393	0.754459	0.680658
+6500	0.282700	0.283174	34.787738	52.455777	0.752326	0.677820
+6750	0.285800	0.283622	34.780553	52.251669	0.754848	0.681448
+7000	0.282800	0.283338	34.545524	52.501134	0.753185	0.678880
+7250	0.285000	0.283650	34.594956	52.309985	0.753798	0.678729
+7500	0.288100	0.283284	34.584192	52.378021	0.753605	0.678208
+7750	0.283700	0.283334	34.755656	52.527052	0.754393	0.679484
+8000	0.288500	0.282929	34.863770	52.089678	0.754177	0.679090
+8250	0.286300	0.283118	35.015418	52.018402	0.755364	0.680887
+8500	0.274100	0.284133	34.665336	52.400700	0.754226	0.679405
+8750	0.266800	0.283913	34.747132	52.335904	0.755003	0.680039
+9000	0.268800	0.284677	34.894368	52.099397	0.756494	0.682901
+9250	0.266900	0.284614	34.798972	52.559451	0.753165	0.678783
+9500	0.261400	0.283952	35.032759	52.099397	0.755384	0.682476
+9750	0.273800	0.284692	34.647750	52.141515	0.752620	0.678391
+10000	0.264200	0.284589	34.573400	52.420139	0.753112	0.679655
+10250	0.264800	0.285057	34.698798	52.556211	0.752907	0.678781
+10500	0.266300	0.285633	34.764845	52.290546	0.754206	0.679841
+10750	0.269000	0.285177	34.831377	52.186872	0.753278	0.679870
+11000	0.268700	0.285041	35.051000	52.122076	0.753716	0.681443
+11250	0.269200	0.285213	34.793976	52.442817	0.752470	0.678178
+11500	0.270100	0.285189	34.979456	52.128556	0.755153	0.680690
+11750	0.268300	0.285230	34.803290	52.459016	0.753642	0.679593
+12000	0.273900	0.285383	34.772534	52.449297	0.753267	0.679359
+12250	0.265500	0.284756	34.976720	52.196592	0.753488	0.680104
+12500	0.268600	0.284991	34.625025	52.711722	0.751160	0.676630
+12750	0.249300	0.285994	34.644483	52.429858	0.752471	0.678399
+13000	0.247200	0.286402	34.472728	52.481695	0.752041	0.677051
+13250	0.253100	0.285756	34.495794	52.378021	0.753062	0.679352
+13500	0.256200	0.286078	34.659761	52.271107	0.754002	0.679929
+13750	0.251700	0.286377	34.811811	52.011923	0.755171	0.681603
+14000	0.252300	0.286590	34.628583	52.180393	0.752986	0.678800
+14250	0.254900	0.285408	34.779966	52.122076	0.753706	0.679666
+14500	0.255900	0.285430	34.523533	52.517333	0.751732	0.676663
+14750	0.256300	0.286176	34.711407	52.559451	0.751216	0.676992
+15000	0.257000	0.285813	34.657526	52.608048	0.751954	0.678795
+15250	0.257300	0.286247	34.357185	52.883432	0.751354	0.678527
+15500	0.256600	0.286103	34.516887	52.407179	0.751949	0.678904
+15750	0.257800	0.285732	34.663972	52.394220	0.752979	0.679021
+16000	0.255700	0.285848	34.811322	52.251669	0.753607	0.681053
+16250	0.251800	0.286037	34.781994	52.238709	0.754066	0.680860
+16500	0.254200	0.286227	34.613614	52.507614	0.753560	0.679909
+16750	0.252700	0.287012	34.593908	52.552971	0.752336	0.678807
+17000	0.237300	0.287675	34.563800	52.436338	0.752185	0.678338
+17250	0.244100	0.287874	34.516444	52.339143	0.752009	0.678697
+17500	0.244100	0.287712	34.716501	52.212791	0.753928	0.680122
+17750	0.245200	0.288067	34.446680	52.572410	0.750473	0.676645
+18000	0.239800	0.288097	34.605238	52.533532	0.752129	0.678389
+18250	0.243700	0.287283	34.812002	52.274347	0.754477	0.681288
+18500	0.244600	0.288129	34.739062	52.193352	0.751838	0.678736
+18750	0.245300	0.287901	34.623221	52.663124	0.752117	0.678288
+19000	0.239600	0.288192	34.536433	52.556211	0.752236	0.678925
+19250	0.247400	0.287871	34.466079	52.549731	0.751615	0.678723
+19500	0.244500	0.288099	34.710327	52.309985	0.753673	0.680532
+19750	0.245100	0.287870	34.630253	52.598328	0.751851	0.679287
+20000	0.243900	0.288536	34.644793	52.484935	0.752480	0.679749
+20250	0.246600	0.287813	34.587763	52.585369	0.752426	0.679073
+20500	0.245200	0.287937	34.681474	52.339143	0.754259	0.680305
+20750	0.241800	0.287583	34.395582	52.591849	0.752626	0.679396
+21000	0.238400	0.288713	34.563898	52.436338	0.751552	0.678410
+21250	0.234200	0.288712	34.483443	52.381261	0.752208	0.678948
+21500	0.232500	0.288679	34.650391	52.290546	0.753767	0.680581
+21750	0.232600	0.288971	34.442432	52.540012	0.751720	0.678481
+22000	0.233400	0.288925	34.569535	52.562690	0.752625	0.678821
+22250	0.238600	0.289514	34.551436	52.381261	0.751440	0.677981
+22500	0.234200	0.289626	34.441588	52.497894	0.751520	0.677585
+22750	0.233900	0.289363	34.593922	52.439578	0.752512	0.679121
+23000	0.232200	0.289463	34.525191	52.481695	0.753027	0.679459
+23250	0.234300	0.289507	34.432297	52.630726	0.752456	0.678347
+23500	0.234600	0.289515	34.576305	52.423378	0.755927	0.681484
+23750	0.235400	0.289634	34.493307	52.361822	0.753436	0.679981
+24000	0.236600	0.289308	34.508890	52.501134	0.752473	0.678877
+24250	0.235800	0.289545	34.454653	52.643686	0.752118	0.678529
+24500	0.235900	0.289539	34.631501	52.400700	0.752927	0.679502
+24750	0.235300	0.289798	34.594740	52.543251	0.752297	0.678988
+25000	0.239200	0.289250	34.487818	52.426618	0.754231	0.680105
+25250	0.223000	0.290312	34.649519	52.540012	0.753576	0.680114
+25500	0.228500	0.290456	34.528229	52.455777	0.754807	0.680595
+25750	0.229200	0.290738	34.640926	52.371542	0.751684	0.678085
+26000	0.228300	0.291240	34.449279	52.552971	0.752228	0.678339
+26250	0.227600	0.291488	34.345595	52.530292	0.752314	0.678304
+26500	0.226800	0.291294	34.311404	52.633966	0.751529	0.677572
+26750	0.225400	0.290639	34.451373	52.546491	0.751783	0.678674
+27000	0.225400	0.291197	34.387727	52.504374	0.751414	0.678172
+27250	0.226500	0.291320	34.485161	52.552971	0.751665	0.678004
+27500	0.226300	0.290772	34.589606	52.459016	0.752079	0.679519
+27750	0.226100	0.291098	34.361585	52.488175	0.752428	0.678358
+28000	0.228800	0.291129	34.326145	52.410419	0.754117	0.680403
+28250	0.228100	0.291153	34.519491	52.400700	0.753561	0.680261
+28500	0.228400	0.290862	34.305193	52.679324	0.752277	0.678830
+28750	0.227300	0.291258	34.467356	52.595088	0.753575	0.679180
+29000	0.232900	0.291031	34.559018	52.468736	0.753631	0.679671
+29250	0.224300	0.291368	34.560449	52.494654	0.753912	0.678687
+29500	0.216400	0.292063	34.517452	52.517333	0.754512	0.679899
+29750	0.219100	0.292350	34.500112	52.523813	0.752534	0.678793
+30000	0.219200	0.292020	34.664747	52.345623	0.754454	0.680843
+30250	0.219400	0.292669	34.488506	52.206311	0.753417	0.679643
+30500	0.220600	0.292113	34.604324	52.452537	0.753472	0.679928
+30750	0.222800	0.292533	34.350358	52.556211	0.752131	0.678221
+31000	0.218700	0.292380	34.486867	52.452537	0.752675	0.678633
+31250	0.222200	0.292313	34.588149	52.439578	0.753715	0.680198
+31500	0.222200	0.292347	34.375924	52.695523	0.752555	0.678755
+31750	0.223300	0.292704	34.617165	52.514093	0.752208	0.679307
+32000	0.221300	0.292315	34.383206	52.523813	0.752533	0.678352
+32250	0.221700	0.292228	34.433439	52.536772	0.752186	0.677860
+32500	0.225100	0.292700	34.333668	52.663124	0.752301	0.678470
+32750	0.225300	0.292077	34.401373	52.666364	0.753154	0.679202
+33000	0.226100	0.292253	34.300853	52.578889	0.751793	0.678611
+33250	0.221700	0.292295	34.337892	52.591849	0.752481	0.678538
+33500	0.220700	0.292421	34.275865	52.637206	0.752148	0.678261
+33750	0.214800	0.292514	34.368292	52.689043	0.751829	0.677967
+34000	0.216600	0.293029	34.329284	52.702002	0.751709	0.677994
+34250	0.215900	0.293012	34.525349	52.556211	0.751295	0.677877
+34500	0.217700	0.292965	34.416318	52.510853	0.751597	0.678574
+34750	0.216000	0.293103	34.362141	52.543251	0.752105	0.678586
+35000	0.219800	0.293020	34.547682	52.410419	0.752500	0.679620
+35250	0.221200	0.293055	34.443104	52.510853	0.752530	0.679353
+35500	0.216000	0.293053	34.273942	52.663124	0.751054	0.677627
+35750	0.216500	0.293292	34.302071	52.536772	0.751861	0.677793
+36000	0.216000	0.293259	34.419504	52.507614	0.751413	0.677874
+36250	0.215300	0.293047	34.386519	52.471976	0.751828	0.677827
+36500	0.216000	0.293291	34.353553	52.533532	0.752023	0.678451
+36750	0.219300	0.293247	34.449875	52.514093	0.751933	0.678079
+37000	0.214600	0.293311	34.373529	52.562690	0.751991	0.678362
+37250	0.217000	0.293330	34.427029	52.608048	0.751084	0.677297
+37500	0.219200	0.293212	34.636159	52.510853	0.751741	0.678528
+37750	0.210000	0.293555	34.521696	52.446057	0.752653	0.679411
+38000	0.210900	0.293689	34.564794	52.569170	0.751889	0.678274
+38250	0.218000	0.293487	34.488335	52.588609	0.752255	0.679032
+38500	0.211200	0.293674	34.479109	52.617767	0.752243	0.679262
+38750	0.217300	0.293676	34.442394	52.501134	0.752282	0.679276
+39000	0.212500	0.293607	34.462774	52.468736	0.752388	0.679002
+39250	0.215600	0.293675	34.379423	52.510853	0.752624	0.679402
+39500	0.211600	0.293567	34.462554	52.514093	0.752719	0.679225
+39750	0.210400	0.293551	34.430552	52.484935	0.752396	0.678942
+40000	0.213100	0.293576	34.437984	52.510853	0.752562	0.678946
+40250	0.216200	0.293517	34.431053	52.565930	0.753121	0.679380
+40500	0.213100	0.293545	34.466043	52.494654	0.752099	0.678525
+40750	0.211100	0.293575	34.501940	52.543251	0.752207	0.678912
+41000	0.218000	0.293595	34.494352	52.488175	0.752383	0.679068
+41250	0.218100	0.293574	34.495723	52.533532	0.752160	0.678757
+41500	0.213100	0.293587	34.509502	52.559451	0.752252	0.678877

model/vi_en/Translation_en_vi_official/train.py

@@ -0,0 +1,183 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+
+SRC = 'en'
+TRG = 'vi'
+KIND = 'official'
+name_model = "Helsinki-NLP/opus-mt-vi-en"
+
+# load dữ liệu
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+list_train = dict()
+list_train[SRC] = load_file(f"../data/train/{SRC}.{SRC}")
+list_train[TRG] = load_file(f"../data/train/{TRG}.{TRG}")
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+list_dev = dict()
+list_dev[SRC] = load_file(f"../data/dev/{SRC}-2012.{SRC}")
+list_dev[TRG] = load_file(f"../data/dev/{TRG}-2012.{TRG}")
+
+
+dataset_train = Dataset.from_dict({"source": list_train[SRC], "target": list_train})
+dataset_test = Dataset.from_dict({"source": list_test[SRC], "target": list_test[TRG]})
+dataset_dev = Dataset.from_dict({"source": list_dev[SRC], "target": list_dev[TRG]})
+
+#tải model:
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+#tạo các phương thức cần thiết:
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+def tokenize_function(examples):
+    inputs = [ex for ex in examples['source']]
+    targets = [ex for ex in examples['target']]
+    model_inputs = tokenizer(inputs, max_length=80, truncation=True, padding="max_length") #80 -> 128
+
+    # Sử dụng tokenizer để mã hóa câu đích
+    with tokenizer.as_target_tokenizer():
+        labels = tokenizer(text_target = targets, max_length=80, truncation=True, padding="max_length") # 80 -> 128
+
+    # Thêm nhãn vào kết quả mã hóa
+    model_inputs["labels"] = labels["input_ids"]
+    
+    return model_inputs
+
+def compute_metrics(eval_pred):
+    predictions, labels = eval_pred
+    
+    predictions = np.array(predictions)
+    predictions[predictions < 0] = 0
+    predictions = predictions.tolist()
+    
+    
+    labels = np.array(labels)
+    labels[labels < 0] = 0
+    labels = labels.tolist()
+    
+    decoded_preds = tokenizer.batch_decode(predictions, skip_special_tokens=True)
+    decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(decoded_preds, [decoded_labels])
+
+    # Tính TER
+    ter = sacrebleu.corpus_ter(decoded_preds, [decoded_labels])
+
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(decoded_preds, [decoded_labels])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(decoded_preds, decoded_labels):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+
+    return {
+        "bleu": bleu.score,
+        "ter": ter.score,
+        "chrf": chrf.score,
+        "rouge1": avg_rouge1,
+        "rougeL": avg_rougeL
+    }
+
+# chuẩn bị dữ liệu:
+tokenized_train = dataset_train.map(tokenize_function, batched=True)
+tokenized_test = dataset_test.map(tokenize_function, batched=True)
+
+data_collator = DataCollatorForSeq2Seq(tokenizer, model=model)
+
+# thiết lập callback
+
+from transformers import TrainerCallback
+
+class CustomStepCallback(TrainerCallback):
+    def __init__(self, steps_interval, action_method):
+        self.steps_interval = steps_interval
+        self.action_method = action_method
+
+    def on_step_end(self, args, state, control, **kwargs):
+        # Thực hiện hành động sau mỗi steps_interval bước
+        if state.global_step % self.steps_interval == 0:
+            self.action_method(state.global_step)
+
+# Định nghĩa phương thức hành động
+def custom_action(step):
+    sentence = {
+        'vi': "Chúng tôi là những bác sĩ xuất sắc và anh ấy cũng vậy, do đó ca phẫu thuật chắc chắn sẽ thành công.",
+        'en': "We are excellent doctors and so is he, so the surgery will definitely be successful."
+    }
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(f"Translated: {pred}")
+
+# Khởi tạo callback với số bước và phương thức hành động
+custom_callback = CustomStepCallback(steps_interval=1000, action_method=custom_action)
+
+# thiết lập huấn luyện
+training_args = Seq2SeqTrainingArguments(
+    output_dir=f"./results-{SRC}-{TRG}-base",
+    evaluation_strategy="steps",
+    eval_steps=250, #100 -> 250 x
+    learning_rate=3e-5, #3 -> 2
+    per_device_train_batch_size=16, #8 -> 16
+    per_device_eval_batch_size=16, #8 -> 16
+    weight_decay=0.01,
+    save_total_limit=3,
+    num_train_epochs=10, # 1 -> 10 x
+    predict_with_generate=True,
+    generation_max_length=50,
+    save_steps=500,
+    logging_dir="./logs",          # Thư mục để lưu logs
+    logging_steps=250,  
+    fp16 = True
+)
+
+trainer = Seq2SeqTrainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_train, #dev -> train x
+    eval_dataset=tokenized_test,
+    tokenizer=tokenizer,
+    data_collator=data_collator,
+    compute_metrics=compute_metrics,
+    callbacks=[custom_callback]
+)
+
+if __name__ == "__main__":
+    # huấn luyện và lưu lại mô hình
+    trainer.train()
+    torch.save(model.state_dict(), f"{SRC}-{TRG}-parameters-{KIND}.pth")

model/vi_en/Translation_en_vi_transformer/log/transformer-scratch.log

@@ -0,0 +1,254 @@
+8.8s 1 Collecting sacrebleu
+8.8s 2 Downloading sacrebleu-2.4.2-py3-none-any.whl.metadata (58 kB)
+8.8s 3 [?25l     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/58.0 kB ? eta -:--:--
+8.9s 4 [?25hCollecting portalocker (from sacrebleu)
+8.9s 5 Downloading portalocker-2.10.1-py3-none-any.whl.metadata (8.5 kB)
+8.9s 6 Requirement already satisfied: regex in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (2023.12.25)
+8.9s 7 Requirement already satisfied: tabulate>=0.8.9 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.9.0)
+8.9s 8 Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (1.26.4)
+8.9s 9 Requirement already satisfied: colorama in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.4.6)
+8.9s 10 Requirement already satisfied: lxml in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (5.2.2)
+9.0s 11 Downloading sacrebleu-2.4.2-py3-none-any.whl (106 kB)
+9.0s 12 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/106.7 kB ? eta -:--:--
+9.0s 13 [?25hDownloading portalocker-2.10.1-py3-none-any.whl (18 kB)
+19.8s 14 Installing collected packages: portalocker, sacrebleu
+20.1s 15 Successfully installed portalocker-2.10.1 sacrebleu-2.4.2
+21.4s 16 Collecting rouge_score
+21.5s 17 Downloading rouge_score-0.1.2.tar.gz (17 kB)
+22.5s 18 Preparing metadata (setup.py) ... [?25l- done
+22.5s 19 [?25hRequirement already satisfied: absl-py in /opt/conda/lib/python3.10/site-packages (from rouge_score) (1.4.0)
+22.5s 20 Requirement already satisfied: nltk in /opt/conda/lib/python3.10/site-packages (from rouge_score) (3.2.4)
+22.5s 21 Requirement already satisfied: numpy in /opt/conda/lib/python3.10/site-packages (from rouge_score) (1.26.4)
+22.5s 22 Requirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge_score) (1.16.0)
+22.6s 23 Building wheels for collected packages: rouge_score
+23.9s 24 Building wheel for rouge_score (setup.py) ... [?25l- \ done
+23.9s 25 [?25h  Created wheel for rouge_score: filename=rouge_score-0.1.2-py3-none-any.whl size=24934 sha256=ed16ec7d76f2c23c099684feb5cb007f77b14876267d0424da52ec34b06ad122
+23.9s 26 Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4
+24.0s 27 Successfully built rouge_score
+34.0s 28 Installing collected packages: rouge_score
+34.2s 29 Successfully installed rouge_score-0.1.2
+35.6s 30 Collecting underthesea
+35.6s 31 Downloading underthesea-6.8.4-py3-none-any.whl.metadata (15 kB)
+35.7s 32 Requirement already satisfied: Click>=6.0 in /opt/conda/lib/python3.10/site-packages (from underthesea) (8.1.7)
+35.7s 33 Collecting python-crfsuite>=0.9.6 (from underthesea)
+35.7s 34 Downloading python_crfsuite-0.9.10-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (4.2 kB)
+35.8s 35 Requirement already satisfied: nltk in /opt/conda/lib/python3.10/site-packages (from underthesea) (3.2.4)
+35.8s 36 Requirement already satisfied: tqdm in /opt/conda/lib/python3.10/site-packages (from underthesea) (4.66.4)
+35.8s 37 Requirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from underthesea) (2.32.3)
+35.8s 38 Requirement already satisfied: joblib in /opt/conda/lib/python3.10/site-packages (from underthesea) (1.4.2)
+35.8s 39 Requirement already satisfied: scikit-learn in /opt/conda/lib/python3.10/site-packages (from underthesea) (1.2.2)
+35.8s 40 Requirement already satisfied: PyYAML in /opt/conda/lib/python3.10/site-packages (from underthesea) (6.0.1)
+35.9s 41 Collecting underthesea-core==1.0.4 (from underthesea)
+35.9s 42 Downloading underthesea_core-1.0.4-cp310-cp310-manylinux2010_x86_64.whl.metadata (1.7 kB)
+35.9s 43 Requirement already satisfied: six in /opt/conda/lib/python3.10/site-packages (from nltk->underthesea) (1.16.0)
+35.9s 44 Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->underthesea) (3.3.2)
+35.9s 45 Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->underthesea) (3.6)
+35.9s 46 Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->underthesea) (1.26.18)
+35.9s 47 Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->underthesea) (2024.7.4)
+36.0s 48 Requirement already satisfied: numpy>=1.17.3 in /opt/conda/lib/python3.10/site-packages (from scikit-learn->underthesea) (1.26.4)
+36.0s 49 Requirement already satisfied: scipy>=1.3.2 in /opt/conda/lib/python3.10/site-packages (from scikit-learn->underthesea) (1.11.4)
+36.0s 50 Requirement already satisfied: threadpoolctl>=2.0.0 in /opt/conda/lib/python3.10/site-packages (from scikit-learn->underthesea) (3.2.0)
+36.1s 51 Downloading underthesea-6.8.4-py3-none-any.whl (20.9 MB)
+36.3s 52 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/20.9 MB ? eta -:--:--
+36.3s 53 [?25hDownloading underthesea_core-1.0.4-cp310-cp310-manylinux2010_x86_64.whl (657 kB)
+36.3s 54 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/657.8 kB ? eta -:--:--
+36.3s 55 [?25hDownloading python_crfsuite-0.9.10-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.1 MB)
+36.4s 56 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/1.1 MB ? eta -:--:--
+46.4s 57 [?25hInstalling collected packages: underthesea-core, python-crfsuite, underthesea
+47.2s 58 Successfully installed python-crfsuite-0.9.10 underthesea-6.8.4 underthesea-core-1.0.4
+47.7s 59 Note: you may need to restart the kernel to use updated packages.
+61.3s 60 /opt/conda/lib/python3.10/site-packages/torch/nn/modules/transformer.py:282: UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.self_attn.batch_first was not True(use batch_first for better inference performance)
+61.3s 61 warnings.warn(f"enable_nested_tensor is True, but self.use_nested_tensor is False because {why_not_sparsity_fast_path}")
+65.3s 62 /opt/conda/lib/python3.10/site-packages/torch/nn/functional.py:5076: UserWarning: Support for mismatched key_padding_mask and attn_mask is deprecated. Use same type for both instead.
+65.3s 63 warnings.warn(
+117.3s 64 Step: 21.322%, Train loss: 8.202, Val loss: 6.663, time: 0p52s
+169.5s 65 Step: 42.644%, Train loss: 7.419, Val loss: 6.478, time: 0p52s
+220.3s 66 Step: 63.966%, Train loss: 7.079, Val loss: 6.394, time: 0p50s
+269.9s 67 Step: 85.288%, Train loss: 6.877, Val loss: 6.215, time: 0p49s
+546.3s 68 Và là là là là là là là là là
+546.3s 69 Epoch: 1, Train loss: 6.769, Val loss: 6.157, Bleu: 0.000, Ter: 115.626, CHRF: 1.976, Rough1: 0.008, RoughL: 0.008, Epoch time: 481.781s
+597.5s 70 Step: 21.322%, Train loss: 6.064, Val loss: 6.093, time: 0p51s
+649.3s 71 Step: 42.644%, Train loss: 6.050, Val loss: 6.030, time: 0p51s
+699.9s 72 Step: 63.966%, Train loss: 6.008, Val loss: 6.075, time: 0p50s
+749.5s 73 Step: 85.288%, Train loss: 5.972, Val loss: 5.976, time: 0p49s
+1014.0s 74 Và bạn là một người .
+1014.0s 75 Epoch: 2, Train loss: 5.941, Val loss: 5.945, Bleu: 0.000, Ter: 117.615, CHRF: 3.267, Rough1: 0.019, RoughL: 0.019, Epoch time: 467.688s
+1065.1s 76 Step: 21.322%, Train loss: 5.758, Val loss: 5.902, time: 0p51s
+1117.1s 77 Step: 42.644%, Train loss: 5.753, Val loss: 5.873, time: 0p51s
+1167.5s 78 Step: 63.966%, Train loss: 5.717, Val loss: 5.907, time: 0p50s
+1217.0s 79 Step: 85.288%, Train loss: 5.685, Val loss: 5.773, time: 0p49s
+1468.9s 80 Và tôi không ?
+1468.9s 81 Epoch: 3, Train loss: 5.659, Val loss: 5.825, Bleu: 0.000, Ter: 118.631, CHRF: 2.901, Rough1: 0.016, RoughL: 0.016, Epoch time: 454.850s
+1519.8s 82 Step: 21.322%, Train loss: 5.536, Val loss: 5.783, time: 0p50s
+1571.3s 83 Step: 42.644%, Train loss: 5.542, Val loss: 5.762, time: 0p51s
+1621.8s 84 Step: 63.966%, Train loss: 5.510, Val loss: 5.735, time: 0p50s
+1671.2s 85 Step: 85.288%, Train loss: 5.485, Val loss: 5.690, time: 0p49s
+1913.1s 86 Và tôi không ?
+1913.1s 87 Epoch: 4, Train loss: 5.463, Val loss: 5.704, Bleu: 0.000, Ter: 117.758, CHRF: 2.755, Rough1: 0.016, RoughL: 0.016, Epoch time: 444.210s
+1964.3s 88 Step: 21.322%, Train loss: 5.369, Val loss: 5.644, time: 0p51s
+2016.1s 89 Step: 42.644%, Train loss: 5.375, Val loss: 5.674, time: 0p51s
+2066.6s 90 Step: 63.966%, Train loss: 5.345, Val loss: 5.619, time: 0p50s
+2116.2s 91 Step: 85.288%, Train loss: 5.319, Val loss: 5.614, time: 0p49s
+2339.6s 92 Và tôi đã làm .
+2339.6s 93 Epoch: 5, Train loss: 5.298, Val loss: 5.578, Bleu: 0.000, Ter: 118.445, CHRF: 2.820, Rough1: 0.016, RoughL: 0.016, Epoch time: 426.542s
+2390.8s 94 Step: 21.322%, Train loss: 5.206, Val loss: 5.571, time: 0p51s
+2442.5s 95 Step: 42.644%, Train loss: 5.215, Val loss: 5.570, time: 0p51s
+2493.1s 96 Step: 63.966%, Train loss: 5.188, Val loss: 5.548, time: 0p50s
+2542.5s 97 Step: 85.288%, Train loss: 5.167, Val loss: 5.516, time: 0p49s
+2799.2s 98 Và tôi đã làm .
+2799.2s 99 Epoch: 6, Train loss: 5.148, Val loss: 5.480, Bleu: 0.000, Ter: 119.614, CHRF: 2.906, Rough1: 0.017, RoughL: 0.017, Epoch time: 459.528s
+2850.1s 100 Step: 21.322%, Train loss: 5.072, Val loss: 5.512, time: 0p50s
+2901.8s 101 Step: 42.644%, Train loss: 5.086, Val loss: 5.491, time: 0p51s
+2952.4s 102 Step: 63.966%, Train loss: 5.064, Val loss: 5.489, time: 0p50s
+3001.8s 103 Step: 85.288%, Train loss: 5.045, Val loss: 5.429, time: 0p49s
+3255.8s 104 Và tôi đã làm .
+3255.8s 105 Epoch: 7, Train loss: 5.027, Val loss: 5.424, Bleu: 0.000, Ter: 119.475, CHRF: 3.032, Rough1: 0.018, RoughL: 0.018, Epoch time: 456.665s
+3306.9s 106 Step: 21.322%, Train loss: 4.956, Val loss: 5.448, time: 0p51s
+3358.5s 107 Step: 42.644%, Train loss: 4.973, Val loss: 5.453, time: 0p51s
+3408.9s 108 Step: 63.966%, Train loss: 4.952, Val loss: 5.432, time: 0p50s
+3458.3s 109 Step: 85.288%, Train loss: 4.932, Val loss: 5.372, time: 0p49s
+3715.1s 110 Và đó là một thứ .
+3715.1s 111 Epoch: 8, Train loss: 4.913, Val loss: 5.338, Bleu: 0.000, Ter: 118.407, CHRF: 3.369, Rough1: 0.020, RoughL: 0.020, Epoch time: 459.261s
+3766.2s 112 Step: 21.322%, Train loss: 4.844, Val loss: 5.393, time: 0p51s
+3817.8s 113 Step: 42.644%, Train loss: 4.866, Val loss: 5.379, time: 0p51s
+3868.3s 114 Step: 63.966%, Train loss: 4.847, Val loss: 5.361, time: 0p50s
+3917.7s 115 Step: 85.288%, Train loss: 4.828, Val loss: 5.310, time: 0p49s
+4179.5s 116 Và nó là một thứ .
+4179.5s 117 Epoch: 9, Train loss: 4.808, Val loss: 5.297, Bleu: 0.000, Ter: 117.319, CHRF: 3.121, Rough1: 0.020, RoughL: 0.020, Epoch time: 464.396s
+4230.5s 118 Step: 21.322%, Train loss: 4.740, Val loss: 5.325, time: 0p51s
+4282.2s 119 Step: 42.644%, Train loss: 4.764, Val loss: 5.358, time: 0p51s
+4332.7s 120 Step: 63.966%, Train loss: 4.747, Val loss: 5.284, time: 0p50s
+4382.1s 121 Step: 85.288%, Train loss: 4.728, Val loss: 5.262, time: 0p49s
+4644.0s 122 Và họ đã làm được .
+4644.0s 123 Epoch: 10, Train loss: 4.711, Val loss: 5.245, Bleu: 0.000, Ter: 118.270, CHRF: 3.367, Rough1: 0.021, RoughL: 0.021, Epoch time: 464.559s
+4695.4s 124 Step: 21.322%, Train loss: 4.644, Val loss: 5.289, time: 0p51s
+4747.4s 125 Step: 42.644%, Train loss: 4.669, Val loss: 5.278, time: 0p52s
+4798.3s 126 Step: 63.966%, Train loss: 4.652, Val loss: 5.230, time: 0p50s
+4848.0s 127 Step: 85.288%, Train loss: 4.634, Val loss: 5.223, time: 0p49s
+5115.0s 128 Và đó là một thứ .
+5115.0s 129 Epoch: 11, Train loss: 4.619, Val loss: 5.221, Bleu: 0.000, Ter: 118.888, CHRF: 3.163, Rough1: 0.021, RoughL: 0.021, Epoch time: 470.965s
+5166.2s 130 Step: 21.322%, Train loss: 4.558, Val loss: 5.254, time: 0p51s
+5218.2s 131 Step: 42.644%, Train loss: 4.581, Val loss: 5.232, time: 0p51s
+5269.0s 132 Step: 63.966%, Train loss: 4.561, Val loss: 5.193, time: 0p50s
+5318.8s 133 Step: 85.288%, Train loss: 4.543, Val loss: 5.176, time: 0p49s
+5585.8s 134 Và họ đã bị mất .
+5585.8s 135 Epoch: 12, Train loss: 4.528, Val loss: 5.181, Bleu: 0.000, Ter: 119.485, CHRF: 3.076, Rough1: 0.020, RoughL: 0.020, Epoch time: 470.803s
+5637.0s 136 Step: 21.322%, Train loss: 4.471, Val loss: 5.214, time: 0p51s
+5689.0s 137 Step: 42.644%, Train loss: 4.491, Val loss: 5.211, time: 0p51s
+5739.7s 138 Step: 63.966%, Train loss: 4.471, Val loss: 5.178, time: 0p50s
+5789.3s 139 Step: 85.288%, Train loss: 4.452, Val loss: 5.144, time: 0p49s
+6056.2s 140 Ch��ng ta đã làm được .
+6056.2s 141 Epoch: 13, Train loss: 4.439, Val loss: 5.160, Bleu: 0.000, Ter: 118.703, CHRF: 3.752, Rough1: 0.025, RoughL: 0.025, Epoch time: 470.417s
+6107.6s 142 Step: 21.322%, Train loss: 4.374, Val loss: 5.166, time: 0p51s
+6159.5s 143 Step: 42.644%, Train loss: 4.394, Val loss: 5.152, time: 0p51s
+6210.2s 144 Step: 63.966%, Train loss: 4.377, Val loss: 5.161, time: 0p50s
+6259.9s 145 Step: 85.288%, Train loss: 4.361, Val loss: 5.125, time: 0p49s
+6523.9s 146 Chúng tôi đã làm được .
+6523.9s 147 Epoch: 14, Train loss: 4.350, Val loss: 5.109, Bleu: 0.000, Ter: 118.007, CHRF: 4.421, Rough1: 0.028, RoughL: 0.028, Epoch time: 467.668s
+6575.3s 148 Step: 21.322%, Train loss: 4.284, Val loss: 5.136, time: 0p51s
+6627.3s 149 Step: 42.644%, Train loss: 4.307, Val loss: 5.158, time: 0p51s
+6677.9s 150 Step: 63.966%, Train loss: 4.290, Val loss: 5.110, time: 0p50s
+6727.5s 151 Step: 85.288%, Train loss: 4.273, Val loss: 5.108, time: 0p49s
+6987.4s 152 Chúng tôi đã bị mất .
+6987.4s 153 Epoch: 15, Train loss: 4.260, Val loss: 5.037, Bleu: 0.000, Ter: 118.338, CHRF: 4.579, Rough1: 0.029, RoughL: 0.029, Epoch time: 463.444s
+7038.5s 154 Step: 21.322%, Train loss: 4.174, Val loss: 5.070, time: 0p51s
+7090.2s 155 Step: 42.644%, Train loss: 4.193, Val loss: 5.096, time: 0p51s
+7140.7s 156 Step: 63.966%, Train loss: 4.170, Val loss: 5.012, time: 0p50s
+7190.2s 157 Step: 85.288%, Train loss: 4.152, Val loss: 5.026, time: 0p49s
+7446.1s 158 Chúng tôi đã bị mất .
+7446.1s 159 Epoch: 16, Train loss: 4.136, Val loss: 4.964, Bleu: 0.000, Ter: 117.853, CHRF: 5.226, Rough1: 0.032, RoughL: 0.032, Epoch time: 458.699s
+7496.9s 160 Step: 21.322%, Train loss: 4.042, Val loss: 4.973, time: 0p50s
+7548.5s 161 Step: 42.644%, Train loss: 4.069, Val loss: 5.065, time: 0p51s
+7598.9s 162 Step: 63.966%, Train loss: 4.044, Val loss: 4.954, time: 0p50s
+7648.3s 163 Step: 85.288%, Train loss: 4.026, Val loss: 4.945, time: 0p49s
+7905.5s 164 Chúng tôi đã bị mất .
+7905.5s 165 Epoch: 17, Train loss: 4.009, Val loss: 4.881, Bleu: 0.000, Ter: 118.015, CHRF: 5.203, Rough1: 0.033, RoughL: 0.033, Epoch time: 459.469s
+7956.3s 166 Step: 21.322%, Train loss: 3.907, Val loss: 4.869, time: 0p50s
+8007.9s 167 Step: 42.644%, Train loss: 3.935, Val loss: 4.983, time: 0p51s
+8058.2s 168 Step: 63.966%, Train loss: 3.908, Val loss: 4.880, time: 0p50s
+8107.4s 169 Step: 85.288%, Train loss: 3.890, Val loss: 4.864, time: 0p49s
+8366.9s 170 Chúng ta đều là người đàn ông .
+8366.9s 171 Epoch: 18, Train loss: 3.870, Val loss: 4.764, Bleu: 0.000, Ter: 118.659, CHRF: 4.797, Rough1: 0.030, RoughL: 0.030, Epoch time: 461.416s
+8417.9s 172 Step: 21.322%, Train loss: 3.771, Val loss: 4.794, time: 0p50s
+8469.7s 173 Step: 42.644%, Train loss: 3.799, Val loss: 4.891, time: 0p51s
+8520.2s 174 Step: 63.966%, Train loss: 3.766, Val loss: 4.858, time: 0p50s
+8569.6s 175 Step: 85.288%, Train loss: 3.745, Val loss: 4.794, time: 0p49s
+8829.3s 176 Chúng ta đều là người đàn ông .
+8829.3s 177 Epoch: 19, Train loss: 3.724, Val loss: 4.686, Bleu: 0.000, Ter: 118.646, CHRF: 4.726, Rough1: 0.030, RoughL: 0.030, Epoch time: 462.371s
+8880.0s 178 Step: 21.322%, Train loss: 3.635, Val loss: 4.744, time: 0p50s
+8931.4s 179 Step: 42.644%, Train loss: 3.660, Val loss: 4.802, time: 0p51s
+8981.8s 180 Step: 63.966%, Train loss: 3.624, Val loss: 4.743, time: 0p50s
+9031.3s 181 Step: 85.288%, Train loss: 3.604, Val loss: 4.700, time: 0p49s
+9292.5s 182 Chúng ta đều là người đàn ông .
+9292.5s 183 Epoch: 20, Train loss: 3.582, Val loss: 4.607, Bleu: 0.000, Ter: 118.962, CHRF: 4.898, Rough1: 0.031, RoughL: 0.031, Epoch time: 463.159s
+9343.4s 184 Step: 21.322%, Train loss: 3.507, Val loss: 4.703, time: 0p50s
+9394.9s 185 Step: 42.644%, Train loss: 3.529, Val loss: 4.722, time: 0p51s
+9445.2s 186 Step: 63.966%, Train loss: 3.494, Val loss: 4.716, time: 0p50s
+9494.5s 187 Step: 85.288%, Train loss: 3.475, Val loss: 4.630, time: 0p49s
+9751.0s 188 Chúng ta đều là người đàn ông .
+9751.0s 189 Epoch: 21, Train loss: 3.452, Val loss: 4.582, Bleu: 0.000, Ter: 118.642, CHRF: 4.834, Rough1: 0.031, RoughL: 0.031, Epoch time: 458.568s
+9801.9s 190 Step: 21.322%, Train loss: 3.386, Val loss: 4.632, time: 0p50s
+9853.5s 191 Step: 42.644%, Train loss: 3.402, Val loss: 4.629, time: 0p51s
+9903.9s 192 Step: 63.966%, Train loss: 3.365, Val loss: 4.662, time: 0p50s
+9953.0s 193 Step: 85.288%, Train loss: 3.348, Val loss: 4.538, time: 0p49s
+10208.3s 194 Chúng ta đều là người Mỹ .
+10208.3s 195 Epoch: 22, Train loss: 3.324, Val loss: 4.534, Bleu: 0.000, Ter: 118.765, CHRF: 5.051, Rough1: 0.032, RoughL: 0.032, Epoch time: 457.212s
+10259.0s 196 Step: 21.322%, Train loss: 3.258, Val loss: 4.573, time: 0p50s
+10310.5s 197 Step: 42.644%, Train loss: 3.274, Val loss: 4.584, time: 0p51s
+10361.2s 198 Step: 63.966%, Train loss: 3.233, Val loss: 4.634, time: 0p50s
+10410.5s 199 Step: 85.288%, Train loss: 3.219, Val loss: 4.516, time: 0p49s
+10659.5s 200 Chúng ta đều là người Mỹ .
+10659.5s 201 Epoch: 23, Train loss: 3.195, Val loss: 4.528, Bleu: 0.000, Ter: 118.971, CHRF: 4.982, Rough1: 0.032, RoughL: 0.032, Epoch time: 451.261s
+10710.3s 202 Step: 21.322%, Train loss: 3.130, Val loss: 4.539, time: 0p50s
+10761.6s 203 Step: 42.644%, Train loss: 3.138, Val loss: 4.533, time: 0p51s
+10811.9s 204 Step: 63.966%, Train loss: 3.097, Val loss: 4.625, time: 0p50s
+10861.4s 205 Step: 85.288%, Train loss: 3.085, Val loss: 4.453, time: 0p49s
+11111.1s 206 Chúng ta đều là người Mỹ .
+11111.1s 207 Epoch: 24, Train loss: 3.062, Val loss: 4.504, Bleu: 0.000, Ter: 118.565, CHRF: 5.193, Rough1: 0.033, RoughL: 0.033, Epoch time: 451.563s
+11161.9s 208 Step: 21.322%, Train loss: 2.990, Val loss: 4.590, time: 0p50s
+11213.2s 209 Step: 42.644%, Train loss: 3.007, Val loss: 4.512, time: 0p51s
+11263.5s 210 Step: 63.966%, Train loss: 2.971, Val loss: 4.550, time: 0p50s
+11312.7s 211 Step: 85.288%, Train loss: 2.957, Val loss: 4.402, time: 0p49s
+11571.9s 212 Chúng ta đều là người Mỹ .
+11571.9s 213 Epoch: 25, Train loss: 2.933, Val loss: 4.477, Bleu: 0.000, Ter: 119.301, CHRF: 4.786, Rough1: 0.031, RoughL: 0.031, Epoch time: 460.783s
+11622.7s 214 Step: 21.322%, Train loss: 2.865, Val loss: 4.517, time: 0p50s
+11674.1s 215 Step: 42.644%, Train loss: 2.879, Val loss: 4.505, time: 0p51s
+11724.4s 216 Step: 63.966%, Train loss: 2.849, Val loss: 4.436, time: 0p50s
+11773.7s 217 Step: 85.288%, Train loss: 2.838, Val loss: 4.359, time: 0p49s
+12032.9s 218 Chúng ta đều là người Anh .
+12032.9s 219 Epoch: 26, Train loss: 2.815, Val loss: 4.459, Bleu: 0.000, Ter: 118.969, CHRF: 4.947, Rough1: 0.033, RoughL: 0.033, Epoch time: 460.999s
+12083.8s 220 Step: 21.322%, Train loss: 2.775, Val loss: 4.382, time: 0p50s
+12135.4s 221 Step: 42.644%, Train loss: 2.771, Val loss: 4.465, time: 0p51s
+12185.9s 222 Step: 63.966%, Train loss: 2.734, Val loss: 4.435, time: 0p50s
+12235.1s 223 Step: 85.288%, Train loss: 2.724, Val loss: 4.336, time: 0p49s
+12485.5s 224 Tất cả đều là bác sĩ ,
+12485.5s 225 Epoch: 27, Train loss: 2.702, Val loss: 4.423, Bleu: 0.000, Ter: 118.807, CHRF: 5.049, Rough1: 0.034, RoughL: 0.034, Epoch time: 452.668s
+12536.7s 226 Step: 21.322%, Train loss: 2.666, Val loss: 4.399, time: 0p51s
+12588.4s 227 Step: 42.644%, Train loss: 2.657, Val loss: 4.356, time: 0p51s
+12638.8s 228 Step: 63.966%, Train loss: 2.615, Val loss: 4.448, time: 0p50s
+12688.2s 229 Step: 85.288%, Train loss: 2.605, Val loss: 4.351, time: 0p49s
+12931.5s 230 Tất cả đều là bác sĩ ,
+12931.5s 231 Epoch: 28, Train loss: 2.588, Val loss: 4.460, Bleu: 0.000, Ter: 119.036, CHRF: 4.603, Rough1: 0.031, RoughL: 0.031, Epoch time: 445.937s
+12982.5s 232 Step: 21.322%, Train loss: 2.553, Val loss: 4.396, time: 0p50s
+13034.0s 233 Step: 42.644%, Train loss: 2.541, Val loss: 4.379, time: 0p51s
+13084.3s 234 Step: 63.966%, Train loss: 2.495, Val loss: 4.457, time: 0p50s
+13133.4s 235 Step: 85.288%, Train loss: 2.486, Val loss: 4.371, time: 0p49s
+13382.4s 236 Tất cả đều là bác sĩ .
+13382.4s 237 Epoch: 29, Train loss: 2.468, Val loss: 4.488, Bleu: 0.000, Ter: 119.455, CHRF: 4.418, Rough1: 0.030, RoughL: 0.030, Epoch time: 450.957s
+13433.7s 238 Step: 21.322%, Train loss: 2.426, Val loss: 4.451, time: 0p51s
+13485.6s 239 Step: 42.644%, Train loss: 2.419, Val loss: 4.389, time: 0p51s
+13536.4s 240 Step: 63.966%, Train loss: 2.378, Val loss: 4.416, time: 0p50s
+13586.0s 241 Step: 85.288%, Train loss: 2.371, Val loss: 4.359, time: 0p49s
+13840.6s 242 Chúng ta đều là bác sĩ .
+13840.6s 243 Epoch: 30, Train loss: 2.356, Val loss: 4.498, Bleu: 0.000, Ter: 119.159, CHRF: 4.405, Rough1: 0.030, RoughL: 0.030, Epoch time: 458.029s
+13841.2s 244 Chúng ta đều là bác sĩ .
+13846.9s 245 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["remove_papermill_header.RemovePapermillHeader"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+13846.9s 246 warn(
+13846.9s 247 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+13847.0s 248 [NbConvertApp] Converting notebook __notebook__.ipynb to notebook
+13847.4s 249 [NbConvertApp] Writing 53494 bytes to __notebook__.ipynb
+13849.0s 250 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["nbconvert.preprocessors.ExtractOutputPreprocessor"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+13849.0s 251 warn(
+13849.0s 252 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+13849.0s 253 [NbConvertApp] Converting notebook __notebook__.ipynb to html
+13850.1s 254 [NbConvertApp] Writing 382051 bytes to __results__.html

model/vi_en/Translation_en_vi_transformer/predict.py

@@ -0,0 +1,287 @@
+from torchtext.data.utils import get_tokenizer
+from torchtext.vocab import build_vocab_from_iterator
+from torch.utils.data import DataLoader
+from typing import Iterable, List
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import torch
+import html
+from underthesea import word_tokenize
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import Transformer
+import math
+from timeit import default_timer as timer
+from torch.nn.utils.rnn import pad_sequence
+from torch.utils.data import Dataset
+import pandas as pd
+
+DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+MAX_LENGTH = 64
+SRC_LANGUAGE = 'en'
+TGT_LANGUAGE = 'vi'
+UNK_IDX, PAD_IDX, BOS_IDX, EOS_IDX = 0, 1, 2, 3
+special_symbols = ['<unk>', '<pad>', '<bos>', '<eos>']
+sentence = {
+    'en': "The patient have a good health",
+    'vi': "Những bệnh nhân này hoàn toàn khỏe mạnh"
+}
+
+token_transform = {}
+vocab_transform = {}
+
+# Tạo tokenizer cho tiếng việt, anh
+english_tokenizer_func = get_tokenizer('spacy', language='en_core_web_sm')
+def vietnamese_tokenizer(text):
+    return word_tokenize(text, format="text").split()[:MAX_LENGTH]
+def english_tokenizer(text):
+    return english_tokenizer_func(text)[:MAX_LENGTH]
+
+token_transform['en'] = english_tokenizer
+token_transform['vi'] = vietnamese_tokenizer
+
+# Tạo vocab
+prefix_vocab = "./model/"
+vocab_transform[SRC_LANGUAGE] = torch.load(prefix_vocab+f'vocab-{SRC_LANGUAGE}.pth')
+vocab_transform[TGT_LANGUAGE] = torch.load(prefix_vocab+f'vocab-{TGT_LANGUAGE}.pth')
+
+# Mô hình
+# Mô-đun trợ giúp thêm mã hóa vị trí vào việc nhúng mã thông báo để giới thiệu khái niệm về thứ tự từ.
+class PositionalEncoding(nn.Module):
+    def __init__(self,
+                 emb_size: int,
+                 dropout: float,
+                 maxlen: int = 100):
+        super(PositionalEncoding, self).__init__()
+        den = torch.exp(- torch.arange(0, emb_size, 2)* math.log(10000) / emb_size)
+        pos = torch.arange(0, maxlen).reshape(maxlen, 1)
+        pos_embedding = torch.zeros((maxlen, emb_size))
+        pos_embedding[:, 0::2] = torch.sin(pos * den)
+        pos_embedding[:, 1::2] = torch.cos(pos * den)
+        pos_embedding = pos_embedding.unsqueeze(-2)
+
+        self.dropout = nn.Dropout(dropout)
+        self.register_buffer('pos_embedding', pos_embedding)
+
+    def forward(self, token_embedding: Tensor):
+        return self.dropout(token_embedding + self.pos_embedding[:token_embedding.size(0), :])
+# Model
+# chuyển token thành vector embedding
+class TokenEmbedding(nn.Module):
+    def __init__(self, vocab_size: int, emb_size):
+        super(TokenEmbedding, self).__init__()
+        self.embedding = nn.Embedding(vocab_size, emb_size)
+        self.emb_size = emb_size
+
+    def forward(self, tokens: Tensor):
+        return self.embedding(tokens.long()) * math.sqrt(self.emb_size)
+
+# Seq2Seq Network
+class Seq2SeqTransformer(nn.Module):
+    def __init__(self,
+                 num_encoder_layers: int,
+                 num_decoder_layers: int,
+                 emb_size: int,
+                 nhead: int,
+                 src_vocab_size: int,
+                 tgt_vocab_size: int,
+                 dim_feedforward: int = 512,
+                 dropout: float = 0.1):
+        super(Seq2SeqTransformer, self).__init__()
+        self.transformer = Transformer(d_model=emb_size,
+                                       nhead=nhead,
+                                       num_encoder_layers=num_encoder_layers,
+                                       num_decoder_layers=num_decoder_layers,
+                                       dim_feedforward=dim_feedforward,
+                                       dropout=dropout)
+        self.generator = nn.Linear(emb_size, tgt_vocab_size)
+        self.src_tok_emb = TokenEmbedding(src_vocab_size, emb_size)
+        self.tgt_tok_emb = TokenEmbedding(tgt_vocab_size, emb_size)
+        self.positional_encoding = PositionalEncoding(
+            emb_size, dropout=dropout)
+
+    def forward(self,
+                src: Tensor,
+                trg: Tensor,
+                src_mask: Tensor,
+                tgt_mask: Tensor,
+                src_padding_mask: Tensor,
+                tgt_padding_mask: Tensor,
+                memory_key_padding_mask: Tensor):
+        src_emb = self.positional_encoding(self.src_tok_emb(src))
+        tgt_emb = self.positional_encoding(self.tgt_tok_emb(trg))
+        outs = self.transformer(src_emb, tgt_emb, src_mask, tgt_mask, None,
+                                src_padding_mask, tgt_padding_mask, memory_key_padding_mask)
+        return self.generator(outs)
+
+    def encode(self, src: Tensor, src_mask: Tensor):
+        return self.transformer.encoder(self.positional_encoding(
+                            self.src_tok_emb(src)), src_mask)
+
+    def decode(self, tgt: Tensor, memory: Tensor, tgt_mask: Tensor):
+        return self.transformer.decoder(self.positional_encoding(
+                          self.tgt_tok_emb(tgt)), memory,
+                          tgt_mask)
+
+# Tạo mặt nạ cho Attention
+def generate_square_subsequent_mask(sz):
+    mask = (torch.triu(torch.ones((sz, sz), device=DEVICE)) == 1).transpose(0, 1)
+    mask = mask.float().masked_fill(mask == 0, float('-inf')).masked_fill(mask == 1, float(0.0))
+    return mask
+# sinh ra output sử dụng greedy
+def greedy_decode(model, src, src_mask, max_len, start_symbol):
+    src = src.to(DEVICE)
+    src_mask = src_mask.to(DEVICE)
+
+    memory = model.encode(src, src_mask)
+    ys = torch.ones(1, 1).fill_(start_symbol).type(torch.long).to(DEVICE)
+    for i in range(max_len-1):
+        memory = memory.to(DEVICE)
+        tgt_mask = (generate_square_subsequent_mask(ys.size(0))
+                    .type(torch.bool)).to(DEVICE)
+        out = model.decode(ys, memory, tgt_mask)
+        out = out.transpose(0, 1)
+        prob = model.generator(out[:, -1])
+        _, next_word = torch.max(prob, dim=1)
+        next_word = next_word.item()
+
+        ys = torch.cat([ys,
+                        torch.ones(1, 1).type_as(src.data).fill_(next_word)], dim=0)
+        if next_word == EOS_IDX:
+            break
+    return ys
+
+def sequential_transforms(*transforms):
+    def func(txt_input):
+        for transform in transforms:
+            txt_input = transform(txt_input)
+        return txt_input
+    return func
+
+# Thêm các kí tự đầu cuối
+def tensor_transform(token_ids: List[int]):
+    return torch.cat((torch.tensor([BOS_IDX]),
+                      torch.tensor(token_ids),
+                      torch.tensor([EOS_IDX])))
+
+# chuyển string thành các chỉ số
+text_transform = {}
+for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
+    text_transform[ln] = sequential_transforms(token_transform[ln], #Tokenization
+                                               vocab_transform[ln], #Numericalization
+                                               tensor_transform) # Add BOS/EOS and create tensor
+# hàm dịch
+def translate(model: torch.nn.Module, src_sentence: str):
+    model.eval()
+
+    src = text_transform[SRC_LANGUAGE](src_sentence).view(-1, 1)
+        
+    num_tokens = src.shape[0]
+    src_mask = (torch.zeros(num_tokens, num_tokens)).type(torch.bool)
+    tgt_tokens = greedy_decode(
+        model,  src, src_mask, max_len=num_tokens + 5, start_symbol=BOS_IDX).flatten()
+    
+    tokens = vocab_transform[TGT_LANGUAGE].lookup_tokens(list(tgt_tokens.cpu().numpy()))
+
+    predict = []
+    for token in tokens:
+        subs_token = token.split("_")
+        if len(subs_token) > 1:
+            predict += subs_token
+        else:
+            predict.append(token)
+
+    predict = [item for item in predict if item not in special_symbols]
+
+    return " ".join(predict).replace("<bos>", "").replace("<eos>", "")
+# Tính toán các metric
+def compute_metrics(decoded_preds, decoded_labels):
+    """
+    input: 
+        decoded_preds: ['A','B','C','D']
+        decoded_labels: ['a','b','c','d']
+    """
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(decoded_preds, [decoded_labels])
+
+    # Tính TER
+    ter = sacrebleu.corpus_ter(decoded_preds, [decoded_labels])
+    
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(decoded_preds, [decoded_labels])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(decoded_preds, decoded_labels):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+
+    return {
+        "bleu": bleu.score,
+        "ter": ter.score,
+        "chrf": chrf.score,
+        "rouge1": avg_rouge1,
+        "rougeL": avg_rougeL
+    }
+
+#Load data
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+if __name__ == "__main__":
+    # load model
+    transformer = torch.load(f"./model/model_{SRC_LANGUAGE}_{TGT_LANGUAGE}.pth", map_location = DEVICE)
+    print(translate(transformer, sentence[SRC_LANGUAGE]))
+
+    prefix_translation = "../data/"
+
+    list_test_en = load_file(prefix_translation+"test/en-2013.en")
+    list_test_vi = load_file(prefix_translation+"test/vi-2013.vi")
+
+    data = {
+        'en':{
+            'test': list_test_en,
+        },
+        'vi':{
+            'test': list_test_vi,
+        }
+    }
+
+    start_time = timer()
+    src = data[SRC_LANGUAGE]['test']
+    reference = data[TGT_LANGUAGE]['test']
+    candidate = [translate(transformer, sent) for sent in src]
+
+    translation = pd.DataFrame({
+        'inputs': src,
+        'preds': candidate,
+        'labels': reference
+    })
+
+    metrics = compute_metrics(candidate, reference)
+
+    result = (f"BLEU: {metrics['bleu']}, TER: {metrics['ter']}, CHRF: {metrics['chrf']}, ROUGE1: {metrics['rouge1']}, ROUGEL: {metrics['rougeL']}")
+
+    f = open('./result/metrics.txt', 'w')
+    f.write(result)
+    f.close()
+
+    translation.to_csv('./result/translation.csv', index=False)
+    end_time = timer()
+    print(f"Save sucessfull after {end_time-start_time}")
+
+

model/vi_en/Translation_en_vi_transformer/result/metrics.txt

@@ -0,0 +1 @@
+BLEU: 11.861815829790698, TER: 72.4630366367793, CHRF: 29.322997294803642, ROUGE1: 0.5952671476070954, ROUGEL: 0.47220229602754843

model/vi_en/Translation_en_vi_transformer/train.py

@@ -0,0 +1,427 @@
+from torchtext.data.utils import get_tokenizer
+from torchtext.vocab import build_vocab_from_iterator
+from torch.utils.data import DataLoader
+from typing import Iterable, List
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import torch
+import html
+from underthesea import word_tokenize
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import Transformer
+import math
+from timeit import default_timer as timer
+from torch.nn.utils.rnn import pad_sequence
+from torch.utils.data import Dataset
+
+DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+MAX_LENGTH = 64
+SRC_LANGUAGE = 'en'
+TGT_LANGUAGE = 'vi'
+#config các tham số và siêu tham số mô hình
+EMB_SIZE = 512
+NHEAD = 8
+FFN_HID_DIM = 2048
+BATCH_SIZE = 64 # 32 -> 128
+NUM_ENCODER_LAYERS = 6
+NUM_DECODER_LAYERS = 6
+sentence = {
+    'en': 'we all are doctors',
+    'vi': "Chúng tôi là những bác sĩ"
+}
+
+#Load data
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+prefix_translation = "../data/"
+
+list_train_en = load_file(prefix_translation+"train/en.en")
+list_train_vi = load_file(prefix_translation+"train/vi.vi")
+
+list_test_en = load_file(prefix_translation+"test/en-2013.en")
+list_test_vi = load_file(prefix_translation+"test/vi-2013.vi")
+
+list_dev_en = load_file(prefix_translation+"dev/en-2012.en")
+list_dev_vi = load_file(prefix_translation+"dev/vi-2012.vi")
+
+data = {
+    'en':{
+        'train': list_train_en,
+        'test': list_test_en,
+        'dev': list_dev_en
+    },
+    'vi':{
+        'train': list_train_vi,
+        'test': list_test_vi,
+        'dev': list_dev_vi
+    }
+}
+
+# Place-holders
+token_transform = {}
+vocab_transform = {}
+
+
+# Tạo tokenizer cho tiếng việt, anh
+english_tokenizer_func = get_tokenizer('spacy', language='en_core_web_sm')
+def vietnamese_tokenizer(text):
+    return word_tokenize(text, format="text").split()[:MAX_LENGTH]
+def english_tokenizer(text):
+    return english_tokenizer_func(text)[:MAX_LENGTH]
+
+token_transform['en'] = english_tokenizer
+token_transform['vi'] = vietnamese_tokenizer
+
+# Xây vocab và lưu lại
+def yield_tokens(data_iter: Iterable, language: str) -> List[str]:
+
+    for data_sample in data_iter[language]['train']:
+        yield token_transform[language](data_sample)
+
+# định nghĩa các kí tự đặc biệt
+UNK_IDX, PAD_IDX, BOS_IDX, EOS_IDX = 0, 1, 2, 3
+
+special_symbols = ['<unk>', '<pad>', '<bos>', '<eos>']
+
+for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
+    # tạo vocab cho từng ngôn ngữ
+    vocab_transform[ln] = build_vocab_from_iterator(yield_tokens(data, ln),
+                                                    min_freq=1,
+                                                    specials=special_symbols,
+                                                    special_first=True)
+for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
+    vocab_transform[ln].set_default_index(UNK_IDX)
+
+torch.save(vocab_transform[SRC_LANGUAGE], f"vocab-{SRC_LANGUAGE}")
+torch.save(vocab_transform[TGT_LANGUAGE], f"vocab-{TGT_LANGUAGE}")
+SRC_VOCAB_SIZE = len(vocab_transform[SRC_LANGUAGE])
+TGT_VOCAB_SIZE = len(vocab_transform[TGT_LANGUAGE])
+
+# Mô hình
+# Mô-đun trợ giúp thêm mã hóa vị trí vào việc nhúng mã thông báo để giới thiệu khái niệm về thứ tự từ.
+class PositionalEncoding(nn.Module):
+    def __init__(self,
+                 emb_size: int,
+                 dropout: float,
+                 maxlen: int = 100):
+        super(PositionalEncoding, self).__init__()
+        den = torch.exp(- torch.arange(0, emb_size, 2)* math.log(10000) / emb_size)
+        pos = torch.arange(0, maxlen).reshape(maxlen, 1)
+        pos_embedding = torch.zeros((maxlen, emb_size))
+        pos_embedding[:, 0::2] = torch.sin(pos * den)
+        pos_embedding[:, 1::2] = torch.cos(pos * den)
+        pos_embedding = pos_embedding.unsqueeze(-2)
+
+        self.dropout = nn.Dropout(dropout)
+        self.register_buffer('pos_embedding', pos_embedding)
+
+    def forward(self, token_embedding: Tensor):
+        return self.dropout(token_embedding + self.pos_embedding[:token_embedding.size(0), :])
+
+# chuyển token thành vector embedding
+class TokenEmbedding(nn.Module):
+    def __init__(self, vocab_size: int, emb_size):
+        super(TokenEmbedding, self).__init__()
+        self.embedding = nn.Embedding(vocab_size, emb_size)
+        self.emb_size = emb_size
+
+    def forward(self, tokens: Tensor):
+        return self.embedding(tokens.long()) * math.sqrt(self.emb_size)
+
+# Seq2Seq Network
+class Seq2SeqTransformer(nn.Module):
+    def __init__(self,
+                 num_encoder_layers: int,
+                 num_decoder_layers: int,
+                 emb_size: int,
+                 nhead: int,
+                 src_vocab_size: int,
+                 tgt_vocab_size: int,
+                 dim_feedforward: int = 512,
+                 dropout: float = 0.1):
+        super(Seq2SeqTransformer, self).__init__()
+        self.transformer = Transformer(d_model=emb_size,
+                                       nhead=nhead,
+                                       num_encoder_layers=num_encoder_layers,
+                                       num_decoder_layers=num_decoder_layers,
+                                       dim_feedforward=dim_feedforward,
+                                       dropout=dropout)
+        self.generator = nn.Linear(emb_size, tgt_vocab_size)
+        self.src_tok_emb = TokenEmbedding(src_vocab_size, emb_size)
+        self.tgt_tok_emb = TokenEmbedding(tgt_vocab_size, emb_size)
+        self.positional_encoding = PositionalEncoding(
+            emb_size, dropout=dropout)
+
+    def forward(self,
+                src: Tensor,
+                trg: Tensor,
+                src_mask: Tensor,
+                tgt_mask: Tensor,
+                src_padding_mask: Tensor,
+                tgt_padding_mask: Tensor,
+                memory_key_padding_mask: Tensor):
+        src_emb = self.positional_encoding(self.src_tok_emb(src))
+        tgt_emb = self.positional_encoding(self.tgt_tok_emb(trg))
+        outs = self.transformer(src_emb, tgt_emb, src_mask, tgt_mask, None,
+                                src_padding_mask, tgt_padding_mask, memory_key_padding_mask)
+        return self.generator(outs)
+
+    def encode(self, src: Tensor, src_mask: Tensor):
+        return self.transformer.encoder(self.positional_encoding(
+                            self.src_tok_emb(src)), src_mask)
+
+    def decode(self, tgt: Tensor, memory: Tensor, tgt_mask: Tensor):
+        return self.transformer.decoder(self.positional_encoding(
+                          self.tgt_tok_emb(tgt)), memory,
+                          tgt_mask)
+
+# Tạo mặt nạ cho Attention
+def generate_square_subsequent_mask(sz):
+    mask = (torch.triu(torch.ones((sz, sz), device=DEVICE)) == 1).transpose(0, 1)
+    mask = mask.float().masked_fill(mask == 0, float('-inf')).masked_fill(mask == 1, float(0.0))
+    return mask
+
+
+def create_mask(src, tgt):
+    src_seq_len = src.shape[0]
+    tgt_seq_len = tgt.shape[0]
+
+    tgt_mask = generate_square_subsequent_mask(tgt_seq_len)
+    src_mask = torch.zeros((src_seq_len, src_seq_len),device=DEVICE).type(torch.bool)
+
+    src_padding_mask = (src == PAD_IDX).transpose(0, 1)
+    tgt_padding_mask = (tgt == PAD_IDX).transpose(0, 1)
+    return src_mask, tgt_mask, src_padding_mask, tgt_padding_mask
+
+# build model
+torch.manual_seed(0)
+
+transformer = Seq2SeqTransformer(NUM_ENCODER_LAYERS, NUM_DECODER_LAYERS, EMB_SIZE,
+                                 NHEAD, SRC_VOCAB_SIZE, TGT_VOCAB_SIZE, FFN_HID_DIM)
+
+for p in transformer.parameters():
+    if p.dim() > 1:
+        nn.init.xavier_uniform_(p)
+
+transformer = transformer.to(DEVICE)
+
+loss_fn = torch.nn.CrossEntropyLoss(ignore_index=PAD_IDX)
+
+optimizer = torch.optim.Adam(transformer.parameters(), lr=0.0001, betas=(0.9, 0.98), eps=1e-9)
+
+#Chuẩn bị dữ liệu
+
+def sequential_transforms(*transforms):
+    def func(txt_input):
+        for transform in transforms:
+            txt_input = transform(txt_input)
+        return txt_input
+    return func
+
+# Thêm các kí tự đầu cuối
+def tensor_transform(token_ids: List[int]):
+    return torch.cat((torch.tensor([BOS_IDX]),
+                      torch.tensor(token_ids),
+                      torch.tensor([EOS_IDX])))
+
+# chuyển string thành các chỉ số
+text_transform = {}
+for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
+    text_transform[ln] = sequential_transforms(token_transform[ln], #Tokenization
+                                               vocab_transform[ln], #Numericalization
+                                               tensor_transform) # Add BOS/EOS and create tensor
+
+
+# chuyển các mẫu dữ liệu vào dataloader
+def collate_fn(batch):
+    src_batch, tgt_batch = [], []
+    for src_sample, tgt_sample in batch:
+        src_batch.append(text_transform[SRC_LANGUAGE](src_sample.rstrip("\n")))
+        tgt_batch.append(text_transform[TGT_LANGUAGE](tgt_sample.rstrip("\n")))
+
+    src_batch = pad_sequence(src_batch, padding_value=PAD_IDX)
+    tgt_batch = pad_sequence(tgt_batch, padding_value=PAD_IDX)
+    return src_batch, tgt_batch
+
+# Tạo dataset
+
+class MyDataset(Dataset):
+    def __init__(self, src, trg):
+        self.src = src
+        self.trg = trg
+    def __len__(self):
+        return len(self.src)
+    
+    def __getitem__(self, index):
+        return self.src[index], self.trg[index]
+train_dataset = MyDataset(data[SRC_LANGUAGE]['train'], data[TGT_LANGUAGE]['train'])
+test_dataset = MyDataset(data[SRC_LANGUAGE]['test'], data[TGT_LANGUAGE]['test'])
+
+# sinh ra output sử dụng greedy
+def greedy_decode(model, src, src_mask, max_len, start_symbol):
+    src = src.to(DEVICE)
+    src_mask = src_mask.to(DEVICE)
+
+    memory = model.encode(src, src_mask)
+    ys = torch.ones(1, 1).fill_(start_symbol).type(torch.long).to(DEVICE)
+    for i in range(max_len-1):
+        memory = memory.to(DEVICE)
+        tgt_mask = (generate_square_subsequent_mask(ys.size(0))
+                    .type(torch.bool)).to(DEVICE)
+        out = model.decode(ys, memory, tgt_mask)
+        out = out.transpose(0, 1)
+        prob = model.generator(out[:, -1])
+        _, next_word = torch.max(prob, dim=1)
+        next_word = next_word.item()
+
+        ys = torch.cat([ys,
+                        torch.ones(1, 1).type_as(src.data).fill_(next_word)], dim=0)
+        if next_word == EOS_IDX:
+            break
+    return ys
+
+
+# hàm dịch
+def translate(model: torch.nn.Module, src_sentence: str):
+    model.eval()
+
+    src = text_transform[SRC_LANGUAGE](src_sentence).view(-1, 1)
+        
+    num_tokens = src.shape[0]
+    src_mask = (torch.zeros(num_tokens, num_tokens)).type(torch.bool)
+    tgt_tokens = greedy_decode(
+        model,  src, src_mask, max_len=num_tokens + 5, start_symbol=BOS_IDX).flatten()
+    
+    tokens = vocab_transform[TGT_LANGUAGE].lookup_tokens(list(tgt_tokens.cpu().numpy()))
+
+    predict = []
+    for token in tokens:
+        subs_token = token.split("_")
+        if len(subs_token) > 1:
+            predict += subs_token
+        else:
+            predict.append(token)
+
+    predict = [item for item in predict if item not in special_symbols]
+
+    return " ".join(predict).replace("<bos>", "").replace("<eos>", "")
+
+def unsqueeze(lst: list, dim = -1):
+    lst = np.expand_dims(lst, axis=dim)
+    return lst.tolist()
+
+def convert_ids_to_sentences(ids: torch.Tensor, language):
+    """
+    input:
+        [[    2,     2,     2,  ...,     2,     2,     2],
+        [ 6693,   133,    12,  ...,    48,   133,   796],
+        [17755,  2869,   177,  ...,    23,   861,     6],
+        ...,
+        [    1,     1,     1,  ...,     1,     1,     1],
+        [    1,     1,     1,  ...,     1,     1,     1],
+        [    1,     1,     1,  ...,     1,     1,     1]]
+    """
+    sentences = []
+    for ids_token in ids.transpose(0,1).tolist():
+        sent = []
+        for token in vocab_transform[language].lookup_tokens(ids_token):
+            subs_token = token.split("_")
+            if len(subs_token) > 1:
+                sent += subs_token
+            else:
+                sent.append(token)
+
+        sent = [item for item in sent if item not in special_symbols]
+        
+        sentences.append(" ".join(sent).replace("<bos>", "").replace("<eos>", "").replace("<pad>",""))
+
+    return sentences
+
+#tạo train và đánh giá cho từng epoch
+def epoch_time(start_time, end_time):
+    elapsed_time = end_time - start_time
+    elapsed_mins = int(elapsed_time / 60)
+    elapsed_secs = int(elapsed_time - (elapsed_mins * 60))
+    return f"{elapsed_mins}p{elapsed_secs}s"
+
+def evaluate(model):
+    model.eval()
+    losses = 0
+
+    val_dataloader = DataLoader(test_dataset, batch_size=BATCH_SIZE, collate_fn=collate_fn)
+
+    for src, tgt in val_dataloader:
+        src = src.to(DEVICE)
+        tgt = tgt.to(DEVICE)
+
+        tgt_input = tgt[:-1, :]
+
+        src_mask, tgt_mask, src_padding_mask, tgt_padding_mask = create_mask(src, tgt_input)
+
+        logits = model(src, tgt_input, src_mask, tgt_mask,src_padding_mask, tgt_padding_mask, src_padding_mask)
+
+        tgt_out = tgt[1:, :]
+        loss = loss_fn(logits.reshape(-1, logits.shape[-1]), tgt_out.reshape(-1))
+        losses += loss.item()
+
+    return losses / len(list(val_dataloader))
+
+def train_epoch(model, optimizer, log_step = 100):
+    losses = []
+    train_dataloader = DataLoader(train_dataset, batch_size=BATCH_SIZE, collate_fn=collate_fn)
+    total = len(train_dataloader)
+    step = 0
+    
+    current_time = timer()
+    for src, tgt in train_dataloader:
+        model.train()
+        src = src.to(DEVICE)
+        tgt = tgt.to(DEVICE)
+
+        tgt_input = tgt[:-1, :]
+
+        src_mask, tgt_mask, src_padding_mask, tgt_padding_mask = create_mask(src, tgt_input)
+
+        logits = model(src, tgt_input, src_mask, tgt_mask,src_padding_mask, tgt_padding_mask, src_padding_mask)
+
+        optimizer.zero_grad()
+
+        tgt_out = tgt[1:, :]
+        loss = loss_fn(logits.reshape(-1, logits.shape[-1]), tgt_out.reshape(-1))
+        loss.backward()
+
+        optimizer.step()
+        losses.append(loss.item())
+        step += 1
+        if step % log_step == 0:
+            val_loss = evaluate(model)
+            end_time = timer()
+            
+            print(f"Step: {(step*100 / total):.3f}%, Train loss: {np.mean(losses):.3f}, Val loss: {val_loss:.3f}, time: {epoch_time(current_time, end_time)}")
+            current_time = timer()
+
+    return np.mean(losses)
+
+#bắt đầu train
+NUM_EPOCHS = 30 # 1 -> 18
+
+for epoch in range(1, NUM_EPOCHS+1):
+    start_time = timer()
+    train_loss = train_epoch(transformer, optimizer) # mở lại
+    val_loss = evaluate(transformer)
+    print(translate(transformer, sentence[SRC_LANGUAGE]))
+    end_time = timer()
+    result = (f"Epoch: {epoch}, Train loss: {train_loss:.3f}, Val loss: {val_loss:.3f}, Epoch time: {(end_time - start_time):.3f}s")
+    print(result)
+
+#Lưu mô hình
+torch.save(transformer, f'model-{SRC_LANGUAGE}-{TGT_LANGUAGE}.pth')

model/vi_en/Translation_vi_en_baseline/final-result/metric.txt

@@ -0,0 +1,2 @@
+bleu	ter	chrf	rouge1	rougeL
+9.771361050393118	85.18033522897336	26.09930283387139	0.36767792667830157	0.34445502776591713

model/vi_en/Translation_vi_en_baseline/log/model-vi-en-baseline.log

@@ -0,0 +1,273 @@
+9.6s 1 Collecting sacrebleu
+9.7s 2 Downloading sacrebleu-2.4.2-py3-none-any.whl.metadata (58 kB)
+9.8s 3 [?25l     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/58.0 kB ? eta -:--:--
+9.9s 4 [?25hCollecting portalocker (from sacrebleu)
+9.9s 5 Downloading portalocker-2.10.1-py3-none-any.whl.metadata (8.5 kB)
+9.9s 6 Requirement already satisfied: regex in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (2023.12.25)
+9.9s 7 Requirement already satisfied: tabulate>=0.8.9 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.9.0)
+9.9s 8 Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (1.26.4)
+9.9s 9 Requirement already satisfied: colorama in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.4.6)
+9.9s 10 Requirement already satisfied: lxml in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (5.2.2)
+10.0s 11 Downloading sacrebleu-2.4.2-py3-none-any.whl (106 kB)
+10.0s 12 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/106.7 kB ? eta -:--:--
+10.1s 13 [?25hDownloading portalocker-2.10.1-py3-none-any.whl (18 kB)
+20.9s 14 Installing collected packages: portalocker, sacrebleu
+21.2s 15 Successfully installed portalocker-2.10.1 sacrebleu-2.4.2
+22.4s 16 Collecting rouge-score
+22.5s 17 Downloading rouge_score-0.1.2.tar.gz (17 kB)
+23.6s 18 Preparing metadata (setup.py) ... [?25l- done
+23.6s 19 [?25hRequirement already satisfied: absl-py in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.4.0)
+23.6s 20 Requirement already satisfied: nltk in /opt/conda/lib/python3.10/site-packages (from rouge-score) (3.2.4)
+23.6s 21 Requirement already satisfied: numpy in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.26.4)
+23.6s 22 Requirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.16.0)
+23.6s 23 Building wheels for collected packages: rouge-score
+25.0s 24 Building wheel for rouge-score (setup.py) ... [?25l- \ done
+25.0s 25 [?25h  Created wheel for rouge-score: filename=rouge_score-0.1.2-py3-none-any.whl size=24934 sha256=3d71c128b31f3c60de3876d6dacad5990c89e8b8d38327d566d012100f87e497
+25.0s 26 Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4
+25.0s 27 Successfully built rouge-score
+35.0s 28 Installing collected packages: rouge-score
+35.2s 29 Successfully installed rouge-score-0.1.2
+35.5s 30 Note: you may need to restart the kernel to use updated packages.
+44.1s 31 2024-08-06 03:37:45.556306: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+44.1s 32 2024-08-06 03:37:45.556419: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+44.1s 33 2024-08-06 03:37:45.736174: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+44.3s 34 2024-08-06 03:37:45.556306: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+44.3s 35 2024-08-06 03:37:45.556419: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+44.3s 36 2024-08-06 03:37:45.736174: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+62.6s 37 /opt/conda/lib/python3.10/site-packages/transformers/tokenization_utils_base.py:4016: UserWarning: `as_target_tokenizer` is deprecated and will be removed in v5 of Transformers. You can tokenize your labels by using the argument `text_target` of the regular `__call__` method (either in the same call as your input texts if you use the same keyword arguments, or in a separate call.
+62.6s 38 warnings.warn(
+88.7s 39 wandb: W&B API key is configured. Use `wandb login --relogin` to force relogin
+88.7s 40 wandb: WARNING If you're specifying your api key in code, ensure this code is not shared publicly.
+88.7s 41 wandb: WARNING Consider setting the WANDB_API_KEY environment variable, or running `wandb login` from the command line.
+88.7s 42 wandb: Appending key for api.wandb.ai to your netrc file: /root/.netrc
+88.9s 43 /opt/conda/lib/python3.10/site-packages/transformers/training_args.py:1494: FutureWarning: `evaluation_strategy` is deprecated and will be removed in version 4.46 of 🤗 Transformers. Use `eval_strategy` instead
+88.9s 44 warnings.warn(
+90.2s 45 wandb: WARNING The `run_name` is currently set to the same value as `TrainingArguments.output_dir`. If this was not intended, please specify a different run name by setting the `TrainingArguments.run_name` parameter.
+90.2s 46 wandb: Currently logged in as: lananhmtms1 (lananhmtms1-mtms1). Use `wandb login --relogin` to force relogin
+105.5s 47 wandb: wandb version 0.17.5 is available!  To upgrade, please run:
+105.5s 48 wandb:  $ pip install wandb --upgrade
+105.5s 49 wandb: Tracking run with wandb version 0.17.4
+105.5s 50 wandb: Run data is saved locally in /kaggle/working/wandb/run-20240806_033832-mrv1507q
+105.5s 51 wandb: Run `wandb offline` to turn off syncing.
+105.5s 52 wandb: Syncing run ./results-vi-en-base
+105.5s 53 wandb: ⭐️ View project at https://wandb.ai/lananhmtms1-mtms1/huggingface
+105.5s 54 wandb: 🚀 View run at https://wandb.ai/lananhmtms1-mtms1/huggingface/runs/mrv1507q
+108.4s 55 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+108.4s 56 warnings.warn('Was asked to gather along dimension 0, but all '
+294.2s 57 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+294.2s 58 warnings.warn('Was asked to gather along dimension 0, but all '
+452.7s 59 Translated: And I think that if I'm a apologist, I'm going to be a pragmatologist.
+485.2s 60 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+485.2s 61 warnings.warn('Was asked to gather along dimension 0, but all '
+675.7s 62 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+675.7s 63 warnings.warn('Was asked to gather along dimension 0, but all '
+834.9s 64 Translated: I mean, if you're going to be able to do that, you're going to be able to do that.
+868.9s 65 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+868.9s 66 warnings.warn('Was asked to gather along dimension 0, but all '
+1062.6s 67 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1062.6s 68 warnings.warn('Was asked to gather along dimension 0, but all '
+1223.5s 69 Translated: And I'm not going to be a sexy, but I'm going to say, that's a sexy, a sexy, a
+1257.7s 70 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1257.7s 71 warnings.warn('Was asked to gather along dimension 0, but all '
+1452.4s 72 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1452.4s 73 warnings.warn('Was asked to gather along dimension 0, but all '
+1612.1s 74 Translated: We have a lot of people who are able to do it and we can do it.
+1647.4s 75 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1647.4s 76 warnings.warn('Was asked to gather along dimension 0, but all '
+1843.2s 77 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1843.2s 78 warnings.warn('Was asked to gather along dimension 0, but all '
+2003.5s 79 Translated: We're talking about the sex and the sex and the sex, and we're going to talk about the sex and the sex.
+2037.8s 80 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2037.8s 81 warnings.warn('Was asked to gather along dimension 0, but all '
+2233.2s 82 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2233.2s 83 warnings.warn('Was asked to gather along dimension 0, but all '
+2394.4s 84 Translated: I'm a sluggish and slamish, and I'm a sluggish slam.
+2429.1s 85 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2429.1s 86 warnings.warn('Was asked to gather along dimension 0, but all '
+2626.0s 87 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2626.0s 88 warnings.warn('Was asked to gather along dimension 0, but all '
+2787.2s 89 Translated: We are a little bit of a symphony and a symphony, which is a very good way to get the symphony to be able to do that
+2823.7s 90 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2823.7s 91 warnings.warn('Was asked to gather along dimension 0, but all '
+3021.2s 92 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3021.2s 93 warnings.warn('Was asked to gather along dimension 0, but all '
+3183.3s 94 Translated: We are the ones that are going to be able to do this and do it, it's a way to make the world more accessible.
+3218.9s 95 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3218.9s 96 warnings.warn('Was asked to gather along dimension 0, but all '
+3416.2s 97 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3416.2s 98 warnings.warn('Was asked to gather along dimension 0, but all '
+3578.6s 99 Translated: We're the symphony and the symboly and the symboly, that's how it's a symphony.
+3614.9s 100 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3614.9s 101 warnings.warn('Was asked to gather along dimension 0, but all '
+3812.5s 102 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3812.5s 103 warnings.warn('Was asked to gather along dimension 0, but all '
+3974.5s 104 Translated: I mean, the sex is a bit more, and it's more of a sex, and it's more of a sex than the sex.
+4011.7s 105 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4011.7s 106 warnings.warn('Was asked to gather along dimension 0, but all '
+4209.2s 107 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4209.2s 108 warnings.warn('Was asked to gather along dimension 0, but all '
+4371.8s 109 Translated: We're the people who are able to eat and eat and eat and drink and they're going to be able to eat.
+4407.4s 110 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4407.4s 111 warnings.warn('Was asked to gather along dimension 0, but all '
+4605.9s 112 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4605.9s 113 warnings.warn('Was asked to gather along dimension 0, but all '
+4767.7s 114 Translated: We are the people who are able to do this and tell them that they're going to be able to do this.
+4804.1s 115 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4804.1s 116 warnings.warn('Was asked to gather along dimension 0, but all '
+5002.3s 117 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5002.3s 118 warnings.warn('Was asked to gather along dimension 0, but all '
+5163.9s 119 Translated: We're a bit more awaited and a little bit more awaited, but we're a little more awaited, and we're a little more awaited,
+5200.8s 120 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5200.8s 121 warnings.warn('Was asked to gather along dimension 0, but all '
+5400.4s 122 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5400.4s 123 warnings.warn('Was asked to gather along dimension 0, but all '
+5562.2s 124 Translated: We are the syroscopic and he's showing you, it's a very simple way to make a difference.
+5599.0s 125 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5599.0s 126 warnings.warn('Was asked to gather along dimension 0, but all '
+5796.8s 127 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5796.8s 128 warnings.warn('Was asked to gather along dimension 0, but all '
+5958.6s 129 Translated: We are a lot of sexy and sexy, but it's not a lot of sexy.
+5994.7s 130 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5994.7s 131 warnings.warn('Was asked to gather along dimension 0, but all '
+6191.8s 132 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6191.8s 133 warnings.warn('Was asked to gather along dimension 0, but all '
+6353.9s 134 Translated: We are the ones that are a bit more than a little bit more and a little bit more than a little bit more.
+6390.5s 135 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6390.5s 136 warnings.warn('Was asked to gather along dimension 0, but all '
+6590.2s 137 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6590.2s 138 warnings.warn('Was asked to gather along dimension 0, but all '
+6753.5s 139 Translated: We are a team of sex and sex, and we're going to have to make sure that the sex is a sex.
+6789.2s 140 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6789.2s 141 warnings.warn('Was asked to gather along dimension 0, but all '
+6988.2s 142 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6988.2s 143 warnings.warn('Was asked to gather along dimension 0, but all '
+7150.8s 144 Translated: We're a lot of sex, and he's a little bit sucked away, so that's not a lot of sex.
+7186.4s 145 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7186.4s 146 warnings.warn('Was asked to gather along dimension 0, but all '
+7385.9s 147 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7385.9s 148 warnings.warn('Was asked to gather along dimension 0, but all '
+7549.1s 149 Translated: We are a kind of a sexy, and we're a little more sexy, and we're a little more sexy.
+7585.0s 150 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7585.0s 151 warnings.warn('Was asked to gather along dimension 0, but all '
+7782.6s 152 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7782.6s 153 warnings.warn('Was asked to gather along dimension 0, but all '
+7944.9s 154 Translated: We're all a little bit, and we're all a little bit, but it's a little bit a little bit sluggish, and it's a
+7981.1s 155 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7981.1s 156 warnings.warn('Was asked to gather along dimension 0, but all '
+8181.5s 157 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8181.5s 158 warnings.warn('Was asked to gather along dimension 0, but all '
+8344.3s 159 Translated: We're all of the sexy and sexy, and you're going to be able to make the sexy, sexy, and that's why we
+8380.4s 160 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8380.4s 161 warnings.warn('Was asked to gather along dimension 0, but all '
+8580.7s 162 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8580.7s 163 warnings.warn('Was asked to gather along dimension 0, but all '
+8744.6s 164 Translated: We were the only people who were interested in and who were interested in that, that they were not afraid to be a part of the society.
+8780.1s 165 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8780.1s 166 warnings.warn('Was asked to gather along dimension 0, but all '
+8978.9s 167 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8978.9s 168 warnings.warn('Was asked to gather along dimension 0, but all '
+9142.8s 169 Translated: We're all the sandboxes and he's, and that's why we're going to be able to make the world a better place.
+9179.6s 170 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9179.6s 171 warnings.warn('Was asked to gather along dimension 0, but all '
+9378.1s 172 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9378.1s 173 warnings.warn('Was asked to gather along dimension 0, but all '
+9541.9s 174 Translated: We're a bunch of sandbags and he's doing that, and that's a very important thing to do.
+9578.0s 175 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9578.0s 176 warnings.warn('Was asked to gather along dimension 0, but all '
+9777.2s 177 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9777.2s 178 warnings.warn('Was asked to gather along dimension 0, but all '
+9939.5s 179 Translated: We're just the sex and the way you see, so that the brain is not going to be a human.
+9975.6s 180 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9975.6s 181 warnings.warn('Was asked to gather along dimension 0, but all '
+10174.1s 182 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10174.1s 183 warnings.warn('Was asked to gather along dimension 0, but all '
+10336.1s 184 Translated: We are the sexy, and he's a sexier, and that's the best way to get a good sex.
+10371.7s 185 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10371.7s 186 warnings.warn('Was asked to gather along dimension 0, but all '
+10569.8s 187 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10569.8s 188 warnings.warn('Was asked to gather along dimension 0, but all '
+10732.4s 189 Translated: We're the ones that are sanding and sanding and letting him go, which are the most important things in the world.
+10769.0s 190 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10769.0s 191 warnings.warn('Was asked to gather along dimension 0, but all '
+10966.2s 192 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10966.2s 193 warnings.warn('Was asked to gather along dimension 0, but all '
+11128.5s 194 Translated: We're the only bifths and he's doing this, and that's the most important part of the equation.
+11164.4s 195 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11164.4s 196 warnings.warn('Was asked to gather along dimension 0, but all '
+11363.5s 197 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11363.5s 198 warnings.warn('Was asked to gather along dimension 0, but all '
+11526.1s 199 Translated: We're the people who've been there and he's here, and that's what we've done for our clients.
+11562.4s 200 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11562.4s 201 warnings.warn('Was asked to gather along dimension 0, but all '
+11759.5s 202 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11759.5s 203 warnings.warn('Was asked to gather along dimension 0, but all '
+11922.0s 204 Translated: We are a bunch of sandbags and he's doing it. We're a bunch of sandbags and sandbags.
+11957.5s 205 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11957.5s 206 warnings.warn('Was asked to gather along dimension 0, but all '
+12154.4s 207 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12154.4s 208 warnings.warn('Was asked to gather along dimension 0, but all '
+12316.8s 209 Translated: We're the people who've been doing this, and we're going to be able to do this.
+12352.2s 210 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12352.2s 211 warnings.warn('Was asked to gather along dimension 0, but all '
+12550.1s 212 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12550.1s 213 warnings.warn('Was asked to gather along dimension 0, but all '
+12713.1s 214 Translated: We're the syringes that he's been able to see, and that's a very good way to see the syringes that he's
+12748.7s 215 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12748.7s 216 warnings.warn('Was asked to gather along dimension 0, but all '
+12946.4s 217 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12946.4s 218 warnings.warn('Was asked to gather along dimension 0, but all '
+13108.5s 219 Translated: We are the sexy and the sexy, and he's doing that, so that the sex of the sexy is a very, very important part of the world
+13144.8s 220 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13144.8s 221 warnings.warn('Was asked to gather along dimension 0, but all '
+13342.4s 222 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13342.4s 223 warnings.warn('Was asked to gather along dimension 0, but all '
+13504.7s 224 Translated: We're the ones who've got the power and the power to do that, which are the most important things to do.
+13540.8s 225 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13540.8s 226 warnings.warn('Was asked to gather along dimension 0, but all '
+13737.9s 227 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13737.9s 228 warnings.warn('Was asked to gather along dimension 0, but all '
+13899.2s 229 Translated: We're a team of scientists and we're able to figure out how to make a better decision.
+13935.2s 230 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13935.2s 231 warnings.warn('Was asked to gather along dimension 0, but all '
+14131.6s 232 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14131.6s 233 warnings.warn('Was asked to gather along dimension 0, but all '
+14293.5s 234 Translated: We are the sex that you're going to be, and you're going to be able to do that.
+14329.0s 235 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14329.0s 236 warnings.warn('Was asked to gather along dimension 0, but all '
+14526.6s 237 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14526.6s 238 warnings.warn('Was asked to gather along dimension 0, but all '
+14688.8s 239 Translated: We're the ones that's going to be, and he's going to be able to do that.
+14724.9s 240 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14724.9s 241 warnings.warn('Was asked to gather along dimension 0, but all '
+14922.3s 242 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14922.3s 243 warnings.warn('Was asked to gather along dimension 0, but all '
+15084.1s 244 Translated: We were the people who've been a little bit, and he was so, so he was a little bit a bit, but he was a little bit, and he was
+15120.0s 245 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15120.0s 246 warnings.warn('Was asked to gather along dimension 0, but all '
+15317.2s 247 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15317.2s 248 warnings.warn('Was asked to gather along dimension 0, but all '
+15479.1s 249 Translated: We are a bunch of sex and sex, and that is why we're not able to make a difference.
+15514.6s 250 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15514.6s 251 warnings.warn('Was asked to gather along dimension 0, but all '
+15712.1s 252 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15712.1s 253 warnings.warn('Was asked to gather along dimension 0, but all '
+15874.2s 254 Translated: We're a group of people who are able to do this, do this by allowing them to do the same thing.
+15909.9s 255 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15909.9s 256 warnings.warn('Was asked to gather along dimension 0, but all '
+16107.7s 257 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16107.7s 258 warnings.warn('Was asked to gather along dimension 0, but all '
+16270.6s 259 Translated: We are a group of scientists and we've seen that this is a re-engineering of the brain.
+16306.1s 260 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16306.1s 261 warnings.warn('Was asked to gather along dimension 0, but all '
+16504.3s 262 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16504.3s 263 warnings.warn('Was asked to gather along dimension 0, but all '
+16554.8s 264 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["remove_papermill_header.RemovePapermillHeader"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+16554.8s 265 warn(
+16554.8s 266 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+16554.8s 267 [NbConvertApp] Converting notebook __notebook__.ipynb to notebook
+16555.3s 268 [NbConvertApp] Writing 201642 bytes to __notebook__.ipynb
+16556.9s 269 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["nbconvert.preprocessors.ExtractOutputPreprocessor"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+16556.9s 270 warn(
+16556.9s 271 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+16557.0s 272 [NbConvertApp] Converting notebook __notebook__.ipynb to html
+16558.0s 273 [NbConvertApp] Writing 458182 bytes to __results__.html

model/vi_en/Translation_vi_en_baseline/predict.py

@@ -0,0 +1,93 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+import pandas as pd
+
+SRC = 'vi'
+TRG = 'en'
+KIND = 'baseline'
+
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+#tải model:
+name_model = f"NguyenManhAI/translation-{SRC}-{TRG}-{KIND}"
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+if __name__ == "__main__":
+    sentence = {
+        'en': "I'm a doctor and so he is",
+        'vi': "Tôi là một bác sĩ và anh ấy cũng vậy."
+    }
+    
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(pred)
+
+    translation = {
+        'inputs':[],
+        'preds':[],
+        'labels':[]
+    }
+
+    for i in range(len(list_test[SRC])):
+        translation['inputs'].append(list_test[SRC][i])
+        translation['preds'].append(predict(model, list_test[SRC][i], tokenizer))
+        translation['labels'].append(list_test[TRG][i])
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(translation['preds'], [translation['labels']])
+    # Tính TER
+    ter = sacrebleu.corpus_ter(translation['preds'], [translation['labels']])
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(translation['preds'], [translation['labels']])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(translation['preds'], translation['labels']):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+    
+    metric_str = f"bleu\tter\tchrf\trouge1\trougeL\n{bleu.score}\t{ter.score}\t{chrf.score}\t{avg_rouge1}\t{avg_rougeL}"
+
+    f = open('final-result\metric.txt', 'w', encoding='utf-8')
+    f.write(metric_str)
+    f.close()
+
+    pd.DataFrame(translation).to_csv('final-result/translation.csv', index=False)
+
+    print("Lưu thành công")

model/vi_en/Translation_vi_en_baseline/result/log-step.csv

@@ -0,0 +1,167 @@
+Step	Training Loss	Validation Loss	Bleu	Ter	Rouge1	Rougel
+250	2.162800	1.096764	0.521002	104.353476	0.097115	0.086380
+500	1.143700	1.064587	0.972787	104.639376	0.117235	0.102468
+750	1.089600	1.050066	1.144112	104.132554	0.126980	0.112194
+1000	1.095500	1.040642	1.328784	104.972926	0.138211	0.122988
+1250	1.077500	1.032666	1.388955	102.269872	0.145518	0.128770
+1500	1.068300	1.026782	1.555567	103.950617	0.146068	0.130340
+1750	1.061400	1.020347	1.634174	102.642408	0.152532	0.136965
+2000	1.059200	1.014594	1.648684	103.326836	0.159372	0.141700
+2250	1.045200	1.009935	1.934424	106.038553	0.164829	0.148806
+2500	1.027800	1.004518	2.106856	104.188867	0.172647	0.157896
+2750	1.044800	0.999143	2.252182	102.889322	0.182297	0.165301
+3000	1.040200	0.994476	2.286156	101.303877	0.186529	0.170626
+3250	1.037100	0.991590	2.424057	102.083604	0.196336	0.179131
+3500	1.015200	0.985237	2.609170	102.157245	0.203719	0.185488
+3750	1.012400	0.981010	2.596388	101.776045	0.204448	0.187336
+4000	1.019500	0.977570	2.742878	104.678363	0.209052	0.191922
+4250	1.025800	0.972629	2.925537	102.287199	0.215092	0.197233
+4500	1.000000	0.968311	3.159274	100.948668	0.222480	0.204713
+4750	1.003700	0.964684	3.263418	99.579814	0.228125	0.210455
+5000	1.010100	0.960824	3.486877	99.995668	0.230582	0.212260
+5250	0.991900	0.958779	3.500702	98.657137	0.231889	0.214064
+5500	0.986200	0.955309	3.823957	99.670782	0.239526	0.220258
+5750	0.992500	0.950842	3.957791	99.649123	0.241434	0.222800
+6000	0.986500	0.948666	4.147657	97.929391	0.247146	0.226243
+6250	0.983100	0.946322	4.200885	100.121291	0.247239	0.227342
+6500	0.976400	0.942853	4.429245	98.691791	0.252030	0.231267
+6750	0.984700	0.940457	4.471030	98.791423	0.252830	0.232217
+7000	0.970700	0.938063	4.639349	99.514836	0.256978	0.235377
+7250	0.977000	0.936263	4.613479	97.418237	0.257306	0.236543
+7500	0.983800	0.932803	4.730498	99.904700	0.257318	0.236944
+7750	0.974600	0.932237	4.689001	97.058696	0.264800	0.242410
+8000	0.982100	0.930422	4.797095	97.556855	0.262845	0.242977
+8250	0.971100	0.926964	4.876241	97.786441	0.264538	0.243591
+8500	0.976700	0.926563	5.062710	96.790123	0.267151	0.245411
+8750	0.960000	0.923590	5.135575	97.370587	0.268811	0.246734
+9000	0.962200	0.922137	5.149504	97.132337	0.271476	0.249216
+9250	0.948600	0.920217	5.158302	97.847087	0.270111	0.248074
+9500	0.946600	0.918721	5.255235	96.664501	0.276053	0.253240
+9750	0.983000	0.917190	5.232128	97.028373	0.270919	0.249986
+10000	0.941000	0.915332	5.371734	97.128005	0.276630	0.254165
+10250	0.940500	0.913624	5.304853	96.248646	0.276222	0.255653
+10500	0.941500	0.911483	5.311787	96.101365	0.275278	0.254983
+10750	0.951100	0.910313	5.406396	96.850769	0.277876	0.256696
+11000	0.950500	0.908519	5.437185	96.136019	0.279347	0.258320
+11250	0.949200	0.906529	5.585474	96.829110	0.280056	0.259370
+11500	0.948500	0.905569	5.678242	96.049383	0.282678	0.261651
+11750	0.949500	0.904321	5.666465	95.183019	0.284296	0.263200
+12000	0.953400	0.902542	5.919460	94.702188	0.286185	0.264932
+12250	0.932400	0.900189	5.983293	95.005415	0.287151	0.265885
+12500	0.935700	0.899032	5.913210	95.611869	0.286351	0.266147
+12750	0.927200	0.898225	6.013556	94.303660	0.289677	0.268971
+13000	0.918400	0.896513	6.100141	94.598224	0.293202	0.270746
+13250	0.937100	0.894367	6.191479	94.758501	0.294896	0.273535
+13500	0.939600	0.892980	6.282865	94.611219	0.295250	0.273358
+13750	0.931300	0.891514	6.394785	94.966428	0.294615	0.272930
+14000	0.927700	0.889602	6.422581	93.753520	0.296687	0.275030
+14250	0.927100	0.889284	6.463763	92.804852	0.298272	0.277021
+14500	0.924200	0.887662	6.468736	93.367988	0.300921	0.278548
+14750	0.936800	0.887166	6.590380	92.869829	0.303396	0.280628
+15000	0.923300	0.885458	6.710997	93.229370	0.302946	0.281368
+15250	0.932800	0.884523	6.659118	92.354343	0.303855	0.282822
+15500	0.926700	0.883579	6.636608	92.982456	0.307451	0.284401
+15750	0.929400	0.882204	6.794384	92.913147	0.307402	0.285422
+16000	0.918200	0.881679	6.828677	93.688542	0.306167	0.285251
+16250	0.911400	0.880150	7.025828	93.597574	0.311014	0.287639
+16500	0.918600	0.879039	6.968754	93.056097	0.308306	0.286065
+16750	0.924700	0.877975	7.028111	92.965129	0.312406	0.290047
+17000	0.910400	0.876075	7.039844	92.869829	0.312418	0.290133
+17250	0.923800	0.875409	7.095285	92.865497	0.313706	0.290889
+17500	0.924600	0.874102	7.095179	93.584579	0.313220	0.290764
+17750	0.919400	0.873187	7.069994	91.734893	0.315221	0.292503
+18000	0.906400	0.873060	7.238890	91.825861	0.317262	0.293862
+18250	0.915500	0.872317	7.372920	93.073424	0.316635	0.293071
+18500	0.913600	0.871288	7.363137	92.332684	0.316608	0.292588
+18750	0.915400	0.870704	7.423542	92.124756	0.319158	0.295239
+19000	0.898400	0.869710	7.353194	91.388347	0.319544	0.296007
+19250	0.916700	0.868732	7.416243	91.124107	0.319153	0.296462
+19500	0.906900	0.868249	7.495710	91.427334	0.321546	0.297136
+19750	0.903300	0.867699	7.519833	92.423652	0.318237	0.296803
+20000	0.900900	0.865547	7.651260	92.570934	0.321265	0.298595
+20250	0.913400	0.865128	7.685165	91.825861	0.323293	0.299880
+20500	0.898300	0.864656	7.600324	91.371020	0.325556	0.302711
+20750	0.893400	0.863313	7.628489	90.760234	0.326127	0.303701
+21000	0.902500	0.862506	7.785776	91.587611	0.325120	0.302070
+21250	0.904100	0.862460	7.859648	91.665584	0.325581	0.302581
+21500	0.903100	0.861502	7.982175	91.193416	0.329159	0.305624
+21750	0.900000	0.861402	7.969100	90.660602	0.328988	0.305583
+22000	0.893800	0.861094	7.911027	89.824561	0.330537	0.307777
+22250	0.920000	0.859362	7.938974	91.392679	0.327932	0.305914
+22500	0.894100	0.859042	7.999565	91.041802	0.328333	0.306446
+22750	0.895000	0.858522	8.048814	91.353693	0.330324	0.307460
+23000	0.887700	0.858366	8.029933	90.578298	0.332386	0.308916
+23250	0.885600	0.858286	8.023738	90.820879	0.331859	0.309072
+23500	0.895100	0.857086	7.911901	90.721248	0.331483	0.308586
+23750	0.891800	0.855909	8.050600	90.916179	0.334184	0.311047
+24000	0.901700	0.855648	8.134091	90.032489	0.334144	0.310884
+24250	0.897100	0.854585	8.099106	90.604288	0.334423	0.311011
+24500	0.896200	0.854505	8.187052	90.305393	0.333490	0.310828
+24750	0.896900	0.854169	8.173603	89.937189	0.335848	0.313233
+25000	0.898900	0.852993	8.120007	90.898852	0.333216	0.310587
+25250	0.888600	0.852599	8.222215	90.465670	0.333940	0.311948
+25500	0.897600	0.852584	8.273651	90.054148	0.336888	0.313601
+25750	0.904500	0.851544	8.262491	90.127789	0.336545	0.314068
+26000	0.883900	0.852164	8.333018	89.590643	0.340138	0.317341
+26250	0.904600	0.850994	8.404498	89.941520	0.339479	0.316496
+26500	0.893500	0.849719	8.425117	90.015161	0.340811	0.317534
+26750	0.879400	0.849673	8.353224	89.482348	0.340142	0.316822
+27000	0.884200	0.849171	8.425437	89.473684	0.342311	0.318182
+27250	0.890100	0.848787	8.562850	89.317739	0.340850	0.317320
+27500	0.877100	0.848616	8.473004	89.685943	0.340169	0.317295
+27750	0.880100	0.848448	8.529262	89.148798	0.343104	0.319841
+28000	0.889100	0.848052	8.436810	89.538661	0.341855	0.318681
+28250	0.879800	0.847456	8.511029	90.123457	0.342883	0.319200
+28500	0.884500	0.846745	8.508162	89.846220	0.341917	0.318477
+28750	0.883400	0.846680	8.532116	89.222439	0.342470	0.319258
+29000	0.904200	0.846338	8.543308	90.127789	0.342719	0.318851
+29250	0.875500	0.846741	8.567820	89.226771	0.342682	0.319640
+29500	0.871600	0.846246	8.602180	89.135802	0.344052	0.321016
+29750	0.885600	0.845252	8.579718	89.400043	0.344634	0.321395
+30000	0.879700	0.844419	8.536548	89.460689	0.344410	0.320758
+30250	0.878500	0.845313	8.536589	89.404375	0.345327	0.321948
+30500	0.883800	0.844427	8.629667	89.088152	0.347419	0.323359
+30750	0.888900	0.843956	8.817481	89.023175	0.346058	0.323177
+31000	0.871800	0.843607	8.654749	89.148798	0.345464	0.321918
+31250	0.883100	0.843170	8.699456	89.707602	0.346530	0.322785
+31500	0.880700	0.842924	8.840148	88.468703	0.347810	0.324866
+31750	0.888900	0.843414	8.848738	88.832575	0.347306	0.324164
+32000	0.879600	0.842126	8.806663	88.992853	0.347291	0.323649
+32250	0.883500	0.842140	8.901572	88.780593	0.348111	0.324213
+32500	0.885000	0.841959	8.831040	88.702621	0.347905	0.324150
+32750	0.885200	0.842284	8.792857	88.780593	0.348569	0.324910
+33000	0.889000	0.841729	8.736697	89.261425	0.346355	0.323071
+33250	0.873100	0.841311	8.883506	89.421702	0.346563	0.323725
+33500	0.877700	0.841128	8.997192	89.443361	0.348613	0.324841
+33750	0.879900	0.840942	8.886926	89.001516	0.349764	0.326112
+34000	0.873400	0.840681	8.918032	88.771930	0.349972	0.325095
+34250	0.879600	0.840439	8.922654	88.940871	0.348729	0.324922
+34500	0.884800	0.839981	8.899336	89.326402	0.348315	0.324284
+34750	0.877100	0.840125	8.932014	88.828243	0.348965	0.325449
+35000	0.889600	0.839630	8.946335	88.512021	0.348552	0.325704
+35250	0.884700	0.839929	8.928207	89.135802	0.346932	0.323808
+35500	0.873300	0.839313	8.966172	89.031839	0.348139	0.324217
+35750	0.883800	0.839204	8.939611	89.174789	0.347672	0.324484
+36000	0.873700	0.839596	9.028072	88.434048	0.350062	0.326299
+36250	0.869000	0.838991	9.046147	88.641975	0.350363	0.327078
+36500	0.879600	0.838835	9.083944	88.317089	0.351520	0.328380
+36750	0.877400	0.838707	9.066646	88.421053	0.350998	0.327455
+37000	0.867300	0.838474	9.031699	88.351744	0.350943	0.327230
+37250	0.868600	0.838576	9.017844	88.481698	0.350245	0.327009
+37500	0.880400	0.838647	9.023358	88.109162	0.351545	0.327698
+37750	0.867200	0.838255	8.994065	88.247780	0.350714	0.326740
+38000	0.871700	0.838353	8.969140	88.546675	0.349989	0.325947
+38250	0.887300	0.838197	9.043189	88.447044	0.350476	0.326348
+38500	0.866900	0.838110	8.966639	88.468703	0.350294	0.325674
+38750	0.881300	0.838204	9.025368	88.295430	0.350948	0.326794
+39000	0.871200	0.837942	9.031211	88.438380	0.350358	0.326644
+39250	0.877800	0.838027	9.041395	88.356075	0.350875	0.327120
+39500	0.858500	0.837755	9.093698	88.139485	0.351147	0.327372
+39750	0.865600	0.837578	9.098669	88.299762	0.351330	0.327553
+40000	0.871600	0.837668	9.093030	88.317089	0.351311	0.327984
+40250	0.879700	0.837648	9.075355	88.334416	0.350948	0.327337
+40500	0.875000	0.837609	9.078352	88.057180	0.351333	0.327603
+40750	0.865000	0.837585	9.083667	87.983539	0.351971	0.328305
+41000	0.883100	0.837574	9.097668	88.009530	0.351911	0.328246
+41250	0.885300	0.837613	9.095008	87.862248	0.352106	0.328451
+41500	0.878700	0.837610	9.096285	87.801603	0.352499	0.328780

model/vi_en/Translation_vi_en_baseline/train.py

@@ -0,0 +1,183 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+
+SRC = 'vi'
+TRG = 'en'
+KIND = 'baseline'
+name_model = "google-t5/t5-small"
+
+# load dữ liệu
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+list_train = dict()
+list_train[SRC] = load_file(f"../data/train/{SRC}.{SRC}")
+list_train[TRG] = load_file(f"../data/train/{TRG}.{TRG}")
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+list_dev = dict()
+list_dev[SRC] = load_file(f"../data/dev/{SRC}-2012.{SRC}")
+list_dev[TRG] = load_file(f"../data/dev/{TRG}-2012.{TRG}")
+
+
+dataset_train = Dataset.from_dict({"source": list_train[SRC], "target": list_train})
+dataset_test = Dataset.from_dict({"source": list_test[SRC], "target": list_test[TRG]})
+dataset_dev = Dataset.from_dict({"source": list_dev[SRC], "target": list_dev[TRG]})
+
+#tải model:
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+#tạo các phương thức cần thiết:
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+def tokenize_function(examples):
+    inputs = [ex for ex in examples['source']]
+    targets = [ex for ex in examples['target']]
+    model_inputs = tokenizer(inputs, max_length=80, truncation=True, padding="max_length") #80 -> 128
+
+    # Sử dụng tokenizer để mã hóa câu đích
+    with tokenizer.as_target_tokenizer():
+        labels = tokenizer(text_target = targets, max_length=80, truncation=True, padding="max_length") # 80 -> 128
+
+    # Thêm nhãn vào kết quả mã hóa
+    model_inputs["labels"] = labels["input_ids"]
+    
+    return model_inputs
+
+def compute_metrics(eval_pred):
+    predictions, labels = eval_pred
+    
+    predictions = np.array(predictions)
+    predictions[predictions < 0] = 0
+    predictions = predictions.tolist()
+    
+    
+    labels = np.array(labels)
+    labels[labels < 0] = 0
+    labels = labels.tolist()
+    
+    decoded_preds = tokenizer.batch_decode(predictions, skip_special_tokens=True)
+    decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(decoded_preds, [decoded_labels])
+
+    # Tính TER
+    ter = sacrebleu.corpus_ter(decoded_preds, [decoded_labels])
+
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(decoded_preds, [decoded_labels])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(decoded_preds, decoded_labels):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+
+    return {
+        "bleu": bleu.score,
+        "ter": ter.score,
+        "chrf": chrf.score,
+        "rouge1": avg_rouge1,
+        "rougeL": avg_rougeL
+    }
+
+# chuẩn bị dữ liệu:
+tokenized_train = dataset_train.map(tokenize_function, batched=True)
+tokenized_test = dataset_test.map(tokenize_function, batched=True)
+
+data_collator = DataCollatorForSeq2Seq(tokenizer, model=model)
+
+# thiết lập callback
+
+from transformers import TrainerCallback
+
+class CustomStepCallback(TrainerCallback):
+    def __init__(self, steps_interval, action_method):
+        self.steps_interval = steps_interval
+        self.action_method = action_method
+
+    def on_step_end(self, args, state, control, **kwargs):
+        # Thực hiện hành động sau mỗi steps_interval bước
+        if state.global_step % self.steps_interval == 0:
+            self.action_method(state.global_step)
+
+# Định nghĩa phương thức hành động
+def custom_action(step):
+    sentence = {
+        'vi': "Chúng tôi là những bác sĩ xuất sắc và anh ấy cũng vậy, do đó ca phẫu thuật chắc chắn sẽ thành công.",
+        'en': "We are excellent doctors and so is he, so the surgery will definitely be successful."
+    }
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(f"Translated: {pred}")
+
+# Khởi tạo callback với số bước và phương thức hành động
+custom_callback = CustomStepCallback(steps_interval=1000, action_method=custom_action)
+
+# thiết lập huấn luyện
+training_args = Seq2SeqTrainingArguments(
+    output_dir=f"./results-{SRC}-{TRG}-base",
+    evaluation_strategy="steps",
+    eval_steps=250, #100 -> 250 x
+    learning_rate=3e-5, #3 -> 2
+    per_device_train_batch_size=16, #8 -> 16
+    per_device_eval_batch_size=16, #8 -> 16
+    weight_decay=0.01,
+    save_total_limit=3,
+    num_train_epochs=10, # 1 -> 10 x
+    predict_with_generate=True,
+    generation_max_length=50,
+    save_steps=500,
+    logging_dir="./logs",          # Thư mục để lưu logs
+    logging_steps=250,  
+    fp16 = True
+)
+
+trainer = Seq2SeqTrainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_train, #dev -> train x
+    eval_dataset=tokenized_test,
+    tokenizer=tokenizer,
+    data_collator=data_collator,
+    compute_metrics=compute_metrics,
+    callbacks=[custom_callback]
+)
+
+if __name__ == "__main__":
+    # huấn luyện và lưu lại mô hình
+    trainer.train()
+    torch.save(model.state_dict(), f"{SRC}-{TRG}-parameters-{KIND}.pth")

model/vi_en/Translation_vi_en_official/final-result/metric.txt

@@ -0,0 +1,2 @@
+bleu	ter	chrf	rouge1	rougeL
+34.560699310846694	60.135438491469614	55.9835775539685	0.6727860801911647	0.6505821121136771

model/vi_en/Translation_vi_en_official/log/model-vi-en-official.log

@@ -0,0 +1,445 @@
+8.1s 1 Collecting sacrebleu
+8.2s 2 Downloading sacrebleu-2.4.2-py3-none-any.whl.metadata (58 kB)
+8.2s 3 [?25l     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/58.0 kB ? eta -:--:--
+8.2s 4 [?25hCollecting portalocker (from sacrebleu)
+8.2s 5 Downloading portalocker-2.10.1-py3-none-any.whl.metadata (8.5 kB)
+8.3s 6 Requirement already satisfied: regex in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (2023.12.25)
+8.3s 7 Requirement already satisfied: tabulate>=0.8.9 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.9.0)
+8.3s 8 Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (1.26.4)
+8.3s 9 Requirement already satisfied: colorama in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (0.4.6)
+8.3s 10 Requirement already satisfied: lxml in /opt/conda/lib/python3.10/site-packages (from sacrebleu) (5.2.2)
+8.3s 11 Downloading sacrebleu-2.4.2-py3-none-any.whl (106 kB)
+8.3s 12 [?25l   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/106.7 kB ? eta -:--:--
+8.3s 13 [?25hDownloading portalocker-2.10.1-py3-none-any.whl (18 kB)
+18.8s 14 Installing collected packages: portalocker, sacrebleu
+19.1s 15 Successfully installed portalocker-2.10.1 sacrebleu-2.4.2
+20.2s 16 Collecting rouge-score
+20.3s 17 Downloading rouge_score-0.1.2.tar.gz (17 kB)
+21.3s 18 Preparing metadata (setup.py) ... [?25l- done
+21.3s 19 [?25hRequirement already satisfied: absl-py in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.4.0)
+21.3s 20 Requirement already satisfied: nltk in /opt/conda/lib/python3.10/site-packages (from rouge-score) (3.2.4)
+21.3s 21 Requirement already satisfied: numpy in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.26.4)
+21.3s 22 Requirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge-score) (1.16.0)
+21.3s 23 Building wheels for collected packages: rouge-score
+22.7s 24 Building wheel for rouge-score (setup.py) ... [?25l- \ done
+22.7s 25 [?25h  Created wheel for rouge-score: filename=rouge_score-0.1.2-py3-none-any.whl size=24934 sha256=33cef78cf4d499c2e86605775b648e5d90bd668a912195b52b03c3d196cb4e10
+22.7s 26 Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4
+22.7s 27 Successfully built rouge-score
+32.6s 28 Installing collected packages: rouge-score
+32.9s 29 Successfully installed rouge-score-0.1.2
+33.2s 30 Note: you may need to restart the kernel to use updated packages.
+40.1s 31 2024-08-05 06:18:30.760235: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+40.1s 32 2024-08-05 06:18:30.760332: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+40.1s 33 2024-08-05 06:18:30.882476: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+40.3s 34 2024-08-05 06:18:30.760235: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+40.3s 35 2024-08-05 06:18:30.760332: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+40.3s 36 2024-08-05 06:18:30.882476: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+54.8s 37 /opt/conda/lib/python3.10/site-packages/torch/_utils.py:831: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
+54.8s 38 return self.fget.__get__(instance, owner)()
+58.6s 39 /opt/conda/lib/python3.10/site-packages/transformers/models/marian/tokenization_marian.py:175: UserWarning: Recommended: pip install sacremoses.
+58.6s 40 warnings.warn("Recommended: pip install sacremoses.")
+60.3s 41 /opt/conda/lib/python3.10/site-packages/transformers/tokenization_utils_base.py:4016: UserWarning: `as_target_tokenizer` is deprecated and will be removed in v5 of Transformers. You can tokenize your labels by using the argument `text_target` of the regular `__call__` method (either in the same call as your input texts if you use the same keyword arguments, or in a separate call.
+60.3s 42 warnings.warn(
+113.7s 43 wandb: W&B API key is configured. Use `wandb login --relogin` to force relogin
+113.7s 44 wandb: WARNING If you're specifying your api key in code, ensure this code is not shared publicly.
+113.7s 45 wandb: WARNING Consider setting the WANDB_API_KEY environment variable, or running `wandb login` from the command line.
+113.7s 46 wandb: Appending key for api.wandb.ai to your netrc file: /root/.netrc
+114.2s 47 /opt/conda/lib/python3.10/site-packages/transformers/training_args.py:1494: FutureWarning: `evaluation_strategy` is deprecated and will be removed in version 4.46 of 🤗 Transformers. Use `eval_strategy` instead
+114.2s 48 warnings.warn(
+115.3s 49 wandb: WARNING The `run_name` is currently set to the same value as `TrainingArguments.output_dir`. If this was not intended, please specify a different run name by setting the `TrainingArguments.run_name` parameter.
+115.3s 50 wandb: Currently logged in as: lananhmtms1 (lananhmtms1-mtms1). Use `wandb login --relogin` to force relogin
+130.9s 51 wandb: wandb version 0.17.5 is available!  To upgrade, please run:
+130.9s 52 wandb:  $ pip install wandb --upgrade
+130.9s 53 wandb: Tracking run with wandb version 0.17.4
+130.9s 54 wandb: Run data is saved locally in /kaggle/working/wandb/run-20240805_061946-6xsf6my3
+130.9s 55 wandb: Run `wandb offline` to turn off syncing.
+130.9s 56 wandb: Syncing run ./results-vi-en
+130.9s 57 wandb: ⭐️ View project at https://wandb.ai/lananhmtms1-mtms1/huggingface
+130.9s 58 wandb: 🚀 View run at https://wandb.ai/lananhmtms1-mtms1/huggingface/runs/6xsf6my3
+133.7s 59 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+133.7s 60 warnings.warn('Was asked to gather along dimension 0, but all '
+463.2s 61 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+463.2s 62 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+465.0s 63 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+465.0s 64 warnings.warn('Was asked to gather along dimension 0, but all '
+704.1s 65 Translated: We're brilliant doctors, and so is he, so the operation is definitely going to work.
+797.8s 66 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+797.8s 67 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+799.6s 68 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+799.6s 69 warnings.warn('Was asked to gather along dimension 0, but all '
+1132.4s 70 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+1132.4s 71 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+1134.3s 72 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1134.3s 73 warnings.warn('Was asked to gather along dimension 0, but all '
+1373.1s 74 Translated: We're great doctors, and so is he, so the surgery is definitely going to work.
+1467.5s 75 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+1467.5s 76 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+1469.4s 77 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1469.4s 78 warnings.warn('Was asked to gather along dimension 0, but all '
+1801.8s 79 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+1801.8s 80 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+1803.7s 81 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+1803.7s 82 warnings.warn('Was asked to gather along dimension 0, but all '
+2043.0s 83 Translated: We're great doctors, and so is he, so the surgery is going to work.
+2137.8s 84 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+2137.8s 85 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+2139.7s 86 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2139.7s 87 warnings.warn('Was asked to gather along dimension 0, but all '
+2473.5s 88 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+2473.5s 89 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+2475.5s 90 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2475.5s 91 warnings.warn('Was asked to gather along dimension 0, but all '
+2713.8s 92 Translated: We're excellent doctors, and so is he, so the surgery is going to work.
+2808.4s 93 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+2808.4s 94 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+2810.4s 95 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+2810.4s 96 warnings.warn('Was asked to gather along dimension 0, but all '
+3143.7s 97 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+3143.7s 98 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+3145.6s 99 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3145.6s 100 warnings.warn('Was asked to gather along dimension 0, but all '
+3385.5s 101 Translated: We're excellent doctors, and so is he, so the operation will definitely succeed.
+3479.2s 102 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+3479.2s 103 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+3481.1s 104 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3481.1s 105 warnings.warn('Was asked to gather along dimension 0, but all '
+3813.5s 106 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+3813.5s 107 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+3815.4s 108 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+3815.4s 109 warnings.warn('Was asked to gather along dimension 0, but all '
+4054.5s 110 Translated: We were excellent doctors, and so was he, so the surgery was going to work.
+4148.7s 111 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+4148.7s 112 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+4150.6s 113 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4150.6s 114 warnings.warn('Was asked to gather along dimension 0, but all '
+4483.7s 115 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+4483.7s 116 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+4485.6s 117 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4485.6s 118 warnings.warn('Was asked to gather along dimension 0, but all '
+4725.0s 119 Translated: We're brilliant doctors, and so is he, so the surgery is certainly going to work.
+4819.3s 120 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+4819.3s 121 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+4821.2s 122 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+4821.2s 123 warnings.warn('Was asked to gather along dimension 0, but all '
+5154.5s 124 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+5154.5s 125 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+5156.4s 126 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5156.4s 127 warnings.warn('Was asked to gather along dimension 0, but all '
+5395.4s 128 Translated: We're excellent doctors, and he's also, so the surgery is going to work.
+5488.8s 129 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+5488.8s 130 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+5490.8s 131 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5490.8s 132 warnings.warn('Was asked to gather along dimension 0, but all '
+5824.4s 133 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+5824.4s 134 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+5826.3s 135 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+5826.3s 136 warnings.warn('Was asked to gather along dimension 0, but all '
+6065.9s 137 Translated: We're excellent doctors, and so is he, so the operation will definitely work.
+6160.2s 138 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+6160.2s 139 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+6162.2s 140 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6162.2s 141 warnings.warn('Was asked to gather along dimension 0, but all '
+6495.1s 142 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+6495.1s 143 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+6497.1s 144 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6497.1s 145 warnings.warn('Was asked to gather along dimension 0, but all '
+6736.8s 146 Translated: We're excellent doctors, and so is he, so the surgery will definitely work.
+6831.1s 147 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+6831.1s 148 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+6833.0s 149 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+6833.0s 150 warnings.warn('Was asked to gather along dimension 0, but all '
+7166.5s 151 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+7166.5s 152 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+7168.4s 153 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7168.4s 154 warnings.warn('Was asked to gather along dimension 0, but all '
+7407.9s 155 Translated: We're brilliant doctors, and so is he, so the operation is going to work.
+7502.1s 156 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+7502.1s 157 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+7504.0s 158 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7504.0s 159 warnings.warn('Was asked to gather along dimension 0, but all '
+7837.5s 160 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+7837.5s 161 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+7839.4s 162 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+7839.4s 163 warnings.warn('Was asked to gather along dimension 0, but all '
+8078.2s 164 Translated: We're brilliant doctors, and he's also, so the surgery will surely work.
+8172.4s 165 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+8172.4s 166 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+8174.3s 167 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8174.3s 168 warnings.warn('Was asked to gather along dimension 0, but all '
+8507.4s 169 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+8507.4s 170 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+8509.3s 171 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8509.3s 172 warnings.warn('Was asked to gather along dimension 0, but all '
+8748.2s 173 Translated: We're excellent doctors, and so is he, so the surgery will definitely work.
+8842.0s 174 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+8842.0s 175 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+8843.9s 176 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+8843.9s 177 warnings.warn('Was asked to gather along dimension 0, but all '
+9176.2s 178 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+9176.2s 179 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+9178.1s 180 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9178.1s 181 warnings.warn('Was asked to gather along dimension 0, but all '
+9417.3s 182 Translated: We are brilliant doctors, and so is he, so the operation is going to work.
+9512.1s 183 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+9512.1s 184 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+9514.0s 185 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9514.0s 186 warnings.warn('Was asked to gather along dimension 0, but all '
+9846.8s 187 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+9846.8s 188 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+9848.7s 189 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+9848.7s 190 warnings.warn('Was asked to gather along dimension 0, but all '
+10087.5s 191 Translated: We were excellent doctors, and so was he, so the surgery was bound to work.
+10182.2s 192 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+10182.2s 193 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+10184.1s 194 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10184.1s 195 warnings.warn('Was asked to gather along dimension 0, but all '
+10517.6s 196 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+10517.6s 197 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+10519.5s 198 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10519.5s 199 warnings.warn('Was asked to gather along dimension 0, but all '
+10758.6s 200 Translated: We're brilliant doctors, and so is he, so the surgery will definitely work.
+10853.5s 201 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+10853.5s 202 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+10855.4s 203 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+10855.4s 204 warnings.warn('Was asked to gather along dimension 0, but all '
+11188.1s 205 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+11188.1s 206 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+11190.1s 207 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11190.1s 208 warnings.warn('Was asked to gather along dimension 0, but all '
+11429.8s 209 Translated: We were brilliant doctors, and so was he, so the surgery was going to work.
+11523.7s 210 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+11523.7s 211 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+11525.6s 212 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11525.6s 213 warnings.warn('Was asked to gather along dimension 0, but all '
+11859.3s 214 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+11859.3s 215 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+11861.2s 216 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+11861.2s 217 warnings.warn('Was asked to gather along dimension 0, but all '
+12101.6s 218 Translated: We're excellent doctors, and so is he, so the surgery will surely succeed.
+12196.0s 219 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+12196.0s 220 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+12197.9s 221 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12197.9s 222 warnings.warn('Was asked to gather along dimension 0, but all '
+12531.6s 223 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+12531.6s 224 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+12533.5s 225 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12533.5s 226 warnings.warn('Was asked to gather along dimension 0, but all '
+12773.0s 227 Translated: We're excellent doctors, and neither is he, so the surgery is going to be successful.
+12867.5s 228 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+12867.5s 229 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+12869.4s 230 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+12869.4s 231 warnings.warn('Was asked to gather along dimension 0, but all '
+13202.8s 232 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+13202.8s 233 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+13204.7s 234 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13204.7s 235 warnings.warn('Was asked to gather along dimension 0, but all '
+13444.6s 236 Translated: We're brilliant doctors, and he's also, so the surgery is certainly going to work.
+13539.2s 237 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+13539.2s 238 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+13541.2s 239 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13541.2s 240 warnings.warn('Was asked to gather along dimension 0, but all '
+13874.0s 241 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+13874.0s 242 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+13875.9s 243 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+13875.9s 244 warnings.warn('Was asked to gather along dimension 0, but all '
+14115.6s 245 Translated: We're excellent doctors, and so is he, so the surgery is certainly going to work.
+14209.7s 246 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+14209.7s 247 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+14211.7s 248 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14211.7s 249 warnings.warn('Was asked to gather along dimension 0, but all '
+14545.1s 250 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+14545.1s 251 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+14547.0s 252 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14547.0s 253 warnings.warn('Was asked to gather along dimension 0, but all '
+14786.0s 254 Translated: We're brilliant doctors, and so is he, so the surgery is going to work.
+14880.6s 255 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+14880.6s 256 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+14882.5s 257 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+14882.5s 258 warnings.warn('Was asked to gather along dimension 0, but all '
+15215.8s 259 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+15215.8s 260 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+15217.7s 261 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15217.7s 262 warnings.warn('Was asked to gather along dimension 0, but all '
+15457.2s 263 Translated: We're very good doctors, and so is he, so the surgery is going to work.
+15551.9s 264 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+15551.9s 265 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+15553.8s 266 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15553.8s 267 warnings.warn('Was asked to gather along dimension 0, but all '
+15886.3s 268 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+15886.3s 269 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+15888.2s 270 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+15888.2s 271 warnings.warn('Was asked to gather along dimension 0, but all '
+16128.0s 272 Translated: We're brilliant doctors, and so is he, so the surgery will probably work.
+16221.8s 273 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+16221.8s 274 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+16223.8s 275 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16223.8s 276 warnings.warn('Was asked to gather along dimension 0, but all '
+16557.2s 277 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+16557.2s 278 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+16559.1s 279 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16559.1s 280 warnings.warn('Was asked to gather along dimension 0, but all '
+16799.3s 281 Translated: We're excellent doctors, and so is he, so the surgery is certainly going to work.
+16893.2s 282 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+16893.2s 283 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+16895.1s 284 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+16895.1s 285 warnings.warn('Was asked to gather along dimension 0, but all '
+17228.3s 286 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+17228.3s 287 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+17230.2s 288 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+17230.2s 289 warnings.warn('Was asked to gather along dimension 0, but all '
+17469.9s 290 Translated: We're excellent doctors, and so is he, so the surgery is definitely going to work.
+17563.8s 291 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+17563.8s 292 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+17565.7s 293 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+17565.7s 294 warnings.warn('Was asked to gather along dimension 0, but all '
+17898.7s 295 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+17898.7s 296 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+17900.7s 297 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+17900.7s 298 warnings.warn('Was asked to gather along dimension 0, but all '
+18139.5s 299 Translated: We're excellent doctors, and so is he, so the operation will surely work.
+18234.1s 300 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+18234.1s 301 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+18236.0s 302 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+18236.0s 303 warnings.warn('Was asked to gather along dimension 0, but all '
+18570.0s 304 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+18570.0s 305 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+18571.9s 306 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+18571.9s 307 warnings.warn('Was asked to gather along dimension 0, but all '
+18811.4s 308 Translated: We're excellent doctors, and so is he, so the operation is bound to be successful.
+18905.3s 309 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+18905.3s 310 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+18907.3s 311 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+18907.3s 312 warnings.warn('Was asked to gather along dimension 0, but all '
+19240.7s 313 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+19240.7s 314 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+19242.7s 315 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+19242.7s 316 warnings.warn('Was asked to gather along dimension 0, but all '
+19482.7s 317 Translated: We're brilliant doctors, and so is he, so the operation will surely work.
+19577.2s 318 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+19577.2s 319 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+19579.1s 320 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+19579.1s 321 warnings.warn('Was asked to gather along dimension 0, but all '
+19912.7s 322 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+19912.7s 323 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+19914.6s 324 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+19914.6s 325 warnings.warn('Was asked to gather along dimension 0, but all '
+20153.9s 326 Translated: We're excellent doctors, and so is he, and so the operation will work.
+20247.9s 327 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+20247.9s 328 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+20249.9s 329 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+20249.9s 330 warnings.warn('Was asked to gather along dimension 0, but all '
+20583.0s 331 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+20583.0s 332 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+20585.0s 333 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+20585.0s 334 warnings.warn('Was asked to gather along dimension 0, but all '
+20824.6s 335 Translated: We're excellent doctors, and so is he, so the operation will surely succeed.
+20918.9s 336 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+20918.9s 337 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+20920.8s 338 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+20920.8s 339 warnings.warn('Was asked to gather along dimension 0, but all '
+21254.3s 340 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+21254.3s 341 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+21256.2s 342 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+21256.2s 343 warnings.warn('Was asked to gather along dimension 0, but all '
+21496.0s 344 Translated: We're excellent doctors, and so is he, so the surgery will work.
+21589.8s 345 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+21589.8s 346 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+21591.7s 347 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+21591.7s 348 warnings.warn('Was asked to gather along dimension 0, but all '
+21925.1s 349 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+21925.1s 350 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+21927.1s 351 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+21927.1s 352 warnings.warn('Was asked to gather along dimension 0, but all '
+22166.0s 353 Translated: We're brilliant doctors, and so is he, so the operation will work.
+22260.3s 354 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+22260.3s 355 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+22262.3s 356 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+22262.3s 357 warnings.warn('Was asked to gather along dimension 0, but all '
+22595.4s 358 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+22595.4s 359 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+22597.4s 360 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+22597.4s 361 warnings.warn('Was asked to gather along dimension 0, but all '
+22836.5s 362 Translated: We were excellent doctors, and so was he, so the surgery was going to work.
+22930.9s 363 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+22930.9s 364 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+22932.9s 365 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+22932.9s 366 warnings.warn('Was asked to gather along dimension 0, but all '
+23266.3s 367 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+23266.3s 368 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+23268.2s 369 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+23268.2s 370 warnings.warn('Was asked to gather along dimension 0, but all '
+23508.3s 371 Translated: We're very good doctors, and so is he, so the operation will definitely work.
+23602.6s 372 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+23602.6s 373 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+23604.5s 374 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+23604.5s 375 warnings.warn('Was asked to gather along dimension 0, but all '
+23937.9s 376 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+23937.9s 377 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+23939.8s 378 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+23939.8s 379 warnings.warn('Was asked to gather along dimension 0, but all '
+24179.4s 380 Translated: We are excellent doctors, and so is he, so the surgery will surely succeed.
+24273.4s 381 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+24273.4s 382 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+24275.3s 383 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+24275.3s 384 warnings.warn('Was asked to gather along dimension 0, but all '
+24608.7s 385 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+24608.7s 386 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+24610.6s 387 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+24610.6s 388 warnings.warn('Was asked to gather along dimension 0, but all '
+24850.4s 389 Translated: We're excellent doctors, and so is he, so the surgery is bound to be a success.
+24945.4s 390 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+24945.4s 391 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+24947.4s 392 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+24947.4s 393 warnings.warn('Was asked to gather along dimension 0, but all '
+25281.8s 394 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+25281.8s 395 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+25283.7s 396 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+25283.7s 397 warnings.warn('Was asked to gather along dimension 0, but all '
+25523.0s 398 Translated: We're excellent doctors, and so is he, so the surgery is going to work.
+25617.7s 399 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+25617.7s 400 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+25619.6s 401 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+25619.6s 402 warnings.warn('Was asked to gather along dimension 0, but all '
+25953.7s 403 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+25953.7s 404 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+25955.6s 405 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+25955.6s 406 warnings.warn('Was asked to gather along dimension 0, but all '
+26195.6s 407 Translated: We're excellent doctors, and so is he, so the surgery is bound to work.
+26289.6s 408 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+26289.6s 409 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+26291.5s 410 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+26291.5s 411 warnings.warn('Was asked to gather along dimension 0, but all '
+26625.3s 412 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+26625.3s 413 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+26627.2s 414 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+26627.2s 415 warnings.warn('Was asked to gather along dimension 0, but all '
+26867.6s 416 Translated: We were excellent doctors, and so was he, so the surgery was bound to succeed.
+26962.2s 417 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+26962.2s 418 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+26964.1s 419 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+26964.1s 420 warnings.warn('Was asked to gather along dimension 0, but all '
+27298.3s 421 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+27298.3s 422 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+27300.2s 423 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+27300.2s 424 warnings.warn('Was asked to gather along dimension 0, but all '
+27540.6s 425 Translated: We're excellent doctors, and so is he, so the surgery will work.
+27634.8s 426 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+27634.8s 427 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+27636.8s 428 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+27636.8s 429 warnings.warn('Was asked to gather along dimension 0, but all '
+27971.2s 430 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+27971.2s 431 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+27973.1s 432 /opt/conda/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.
+27973.1s 433 warnings.warn('Was asked to gather along dimension 0, but all '
+28022.0s 434 Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.
+28022.0s 435 Non-default generation parameters: {'max_length': 512, 'num_beams': 6, 'bad_words_ids': [[53738]], 'forced_eos_token_id': 0}
+28029.5s 436 wandb: - 0.004 MB of 0.004 MB uploaded
+28029.5s 437 warn(
+28029.5s 438 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+28029.6s 439 [NbConvertApp] Converting notebook __notebook__.ipynb to notebook
+28030.2s 440 [NbConvertApp] Writing 249158 bytes to __notebook__.ipynb
+28031.8s 441 /opt/conda/lib/python3.10/site-packages/traitlets/traitlets.py:2930: FutureWarning: --Exporter.preprocessors=["nbconvert.preprocessors.ExtractOutputPreprocessor"] for containers is deprecated in traitlets 5.0. You can pass `--Exporter.preprocessors item` ... multiple times to add items to a list.
+28031.8s 442 warn(
+28031.8s 443 [NbConvertApp] WARNING | Config option `kernel_spec_manager_class` not recognized by `NbConvertApp`.
+28031.8s 444 [NbConvertApp] Converting notebook __notebook__.ipynb to html
+28032.8s 445 [NbConvertApp] Writing 503311 bytes to __results__.html

model/vi_en/Translation_vi_en_official/predict.py

@@ -0,0 +1,93 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, Seq2SeqTrainer, Seq2SeqTrainingArguments, DataCollatorForSeq2Seq
+import sacrebleu
+from rouge_score import rouge_scorer
+import numpy as np
+import html
+from datasets import Dataset
+import pandas as pd
+
+SRC = 'vi'
+TRG = 'en'
+KIND = 'official'
+
+def load_file(file_path):
+    lang = []
+
+    with open(file_path, "r", encoding="utf-8") as file:
+        content_en = file.read()
+    lang += content_en.split('\n')
+    lang = [html.unescape(sent) for sent in lang]
+    return lang
+
+#tải model:
+name_model = f"NguyenManhAI/translation-{SRC}-{TRG}-{KIND}"
+model = AutoModelForSeq2SeqLM.from_pretrained(name_model)
+tokenizer = AutoTokenizer.from_pretrained(name_model)
+
+list_test = dict()
+list_test[SRC] = load_file(f"../data/test/{SRC}-2013.{SRC}")
+list_test[TRG] = load_file(f"../data/test/{TRG}-2013.{TRG}")
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+def predict(model, input_sentence, tokenizer):
+    # Dịch một câu hoàn chỉnh
+    # Token hóa câu đầu vào
+    inputs = tokenizer(input_sentence, return_tensors="pt", padding=True, truncation=True).to(device)
+
+    # Sử dụng mô hình để dự đoán
+    with torch.no_grad():
+        outputs = model.generate(**inputs, max_new_tokens = 50)
+
+    # Giải mã đầu ra của mô hình
+    translated_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return translated_sentence
+
+if __name__ == "__main__":
+    sentence = {
+        'en': "I'm a doctor and so he is",
+        'vi': "Tôi là một bác sĩ và anh ấy cũng vậy."
+    }
+    
+    pred = predict(model, sentence[SRC], tokenizer)
+    print(pred)
+
+    translation = {
+        'inputs':[],
+        'preds':[],
+        'labels':[]
+    }
+
+    for i in range(len(list_test[SRC])):
+        translation['inputs'].append(list_test[SRC][i])
+        translation['preds'].append(predict(model, list_test[SRC][i], tokenizer))
+        translation['labels'].append(list_test[TRG][i])
+
+    # Tính BLEU
+    bleu = sacrebleu.corpus_bleu(translation['preds'], [translation['labels']])
+    # Tính TER
+    ter = sacrebleu.corpus_ter(translation['preds'], [translation['labels']])
+    # Tính CHRF
+    chrf = sacrebleu.corpus_chrf(translation['preds'], [translation['labels']])
+
+    # Tính ROUGE
+    scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
+    rouge1_scores = []
+    rougeL_scores = []
+    for pred, label in zip(translation['preds'], translation['labels']):
+        scores = scorer.score(pred, label)
+        rouge1_scores.append(scores['rouge1'].fmeasure)
+        rougeL_scores.append(scores['rougeL'].fmeasure)
+
+    avg_rouge1 = sum(rouge1_scores) / len(rouge1_scores)
+    avg_rougeL = sum(rougeL_scores) / len(rougeL_scores)
+    
+    metric_str = f"bleu\tter\tchrf\trouge1\trougeL\n{bleu.score}\t{ter.score}\t{chrf.score}\t{avg_rouge1}\t{avg_rougeL}"
+
+    f = open('final-result\metric.txt', 'w', encoding='utf-8')
+    f.write(metric_str)
+    f.close()
+
+    pd.DataFrame(translation).to_csv('final-result/translation.csv', index=False)
+
+    print("Lưu thành công")