Create app.py

app.py

@@ -0,0 +1,104 @@
+import gradio as gr
+import spaces
+from huggingface_hub import hf_hub_download
+from llama_cpp import Llama
+
+MAX_INPUT_LIMIT = 3584
+MAX_NEW_TOKENS = 1536
+MODEL_REPO = "Azure99/blossom-v5.1-34b-gguf"
+MODEL_FILE = "model-q6_k.gguf"
+MODEL_LOCAL_DIR = "./"
+
+hf_hub_download(
+    repo_id=MODEL_REPO,
+    filename=MODEL_FILE,
+    local_dir=MODEL_LOCAL_DIR
+)
+
+llm: Llama = None
+
+
+def get_input_ids(input, history):
+    prefix = "A chat between a human and an artificial intelligence bot. The bot gives helpful, detailed, and polite answers to the human's questions."
+    patterns = []
+    for conv in history:
+        patterns.append(f'\n|Human|: {conv[0]}\n|Bot|: ')
+        patterns.append(f'{conv[1]}')
+    patterns.append(f'\n|Human|: {input}\n|Bot|: ')
+    patterns[0] = prefix + patterns[0]
+
+    input_ids = []
+    for i, pattern in enumerate(patterns):
+        input_ids += llm.tokenize(pattern.encode("utf-8"), add_bos=False, special=True)
+        if i % 2 == 1:
+            input_ids += [llm.token_eos()]
+    return input_ids
+
+
+@spaces.GPU
+def chat(inst, history, temperature, top_p, repetition_penalty):
+    global llm
+    if llm is None:
+        llm = Llama(model_path=MODEL_FILE, n_gpu_layers=-1, flash_attn=True, offload_kqv=True, n_ctx=4096)
+
+    input_ids = get_input_ids(inst, history)
+    if len(input_ids) > MAX_INPUT_LIMIT:
+        yield "The input is too long, please clear the history."
+        return
+
+    generate_config = dict(temperature=temperature, top_p=top_p, repeat_penalty=repetition_penalty,
+                           top_k=50, stream=True, max_tokens=1024)
+
+    outputs = ""
+    for chunk in llm(input_ids, **generate_config):
+        outputs += chunk["choices"][0]["text"]
+        yield outputs
+
+
+additional_inputs = [
+    gr.Slider(
+        label="Temperature",
+        value=0.5,
+        minimum=0.0,
+        maximum=1.0,
+        step=0.05,
+        interactive=True,
+        info="Controls randomness in choosing words.",
+    ),
+    gr.Slider(
+        label="Top-P",
+        value=0.85,
+        minimum=0.0,
+        maximum=1.0,
+        step=0.05,
+        interactive=True,
+        info="Picks words until their combined probability is at least top_p.",
+    ),
+    gr.Slider(
+        label="Repetition penalty",
+        value=1.05,
+        minimum=1.0,
+        maximum=1.2,
+        step=0.01,
+        interactive=True,
+        info="Repetition Penalty: Controls how much repetition is penalized.",
+    )
+]
+
+gr.ChatInterface(chat,
+                 chatbot=gr.Chatbot(show_label=False, height=500, show_copy_button=True, render_markdown=True),
+                 textbox=gr.Textbox(placeholder="", container=False, scale=7),
+                 title="Blossom 34B Demo",
+                 description='Hello, I am Blossom, an open source conversational large language model.🌠'
+                             '<a href="https://github.com/Azure99/BlossomLM">GitHub</a>',
+                 theme="soft",
+                 examples=[["Hello"], ["What is MBTI"], ["用Python实现二分查找"],
+                           ["为switch写一篇小红书种草文案，带上emoji"]],
+                 cache_examples=False,
+                 additional_inputs=additional_inputs,
+                 additional_inputs_accordion=gr.Accordion(label="Config", open=True),
+                 clear_btn="🗑️Clear",
+                 undo_btn="↩️Undo",
+                 retry_btn="🔄Retry",
+                 submit_btn="➡️Submit",
+                 ).queue().launch()