Update README.md
Browse files
README.md
CHANGED
@@ -92,31 +92,7 @@ pip install transformers sentencepiece torch ctranslate2
|
|
92 |
```
|
93 |
|
94 |
### 2. Inference
|
95 |
-
## Vanilla model
|
96 |
-
```Python
|
97 |
-
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
98 |
|
99 |
-
model_path = "anzorq/m2m100_418M_ft_ru-kbd_44K"
|
100 |
-
tgt_lang="zu"
|
101 |
-
|
102 |
-
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
103 |
-
model = AutoModelForSeq2SeqLM.from_pretrained(model_path)
|
104 |
-
|
105 |
-
def translate(text, num_beams=4, num_return_sequences=4):
|
106 |
-
inputs = tokenizer(text, return_tensors="pt")
|
107 |
-
num_return_sequences = min(num_return_sequences, num_beams)
|
108 |
-
|
109 |
-
translated_tokens = model.generate(
|
110 |
-
**inputs, forced_bos_token_id=tokenizer.lang_code_to_id[tgt_lang], num_beams=num_beams, num_return_sequences=num_return_sequences
|
111 |
-
)
|
112 |
-
|
113 |
-
translations = [tokenizer.decode(translation, skip_special_tokens=True) for translation in translated_tokens]
|
114 |
-
return text, translations
|
115 |
-
|
116 |
-
# Test the translation
|
117 |
-
text = "Текст для перевода"
|
118 |
-
print(translate(text))
|
119 |
-
```
|
120 |
|
121 |
## CTranslate2 model (quantized model, much faster inference)
|
122 |
First, download the files for the model in ctranslate2 format:
|
@@ -158,6 +134,32 @@ def translate(text, num_beams=4, num_return_sequences=4):
|
|
158 |
|
159 |
return text, translations
|
160 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
161 |
# Test the translation
|
162 |
text = "Текст для перевода"
|
163 |
print(translate(text))
|
|
|
92 |
```
|
93 |
|
94 |
### 2. Inference
|
|
|
|
|
|
|
95 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
|
97 |
## CTranslate2 model (quantized model, much faster inference)
|
98 |
First, download the files for the model in ctranslate2 format:
|
|
|
134 |
|
135 |
return text, translations
|
136 |
|
137 |
+
# Test the translation
|
138 |
+
text = "Текст для перевода"
|
139 |
+
print(translate(text))
|
140 |
+
```
|
141 |
+
|
142 |
+
## Vanilla model
|
143 |
+
```Python
|
144 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
145 |
+
|
146 |
+
model_path = "anzorq/m2m100_418M_ft_ru-kbd_44K"
|
147 |
+
tgt_lang="zu"
|
148 |
+
|
149 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
150 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(model_path)
|
151 |
+
|
152 |
+
def translate(text, num_beams=4, num_return_sequences=4):
|
153 |
+
inputs = tokenizer(text, return_tensors="pt")
|
154 |
+
num_return_sequences = min(num_return_sequences, num_beams)
|
155 |
+
|
156 |
+
translated_tokens = model.generate(
|
157 |
+
**inputs, forced_bos_token_id=tokenizer.lang_code_to_id[tgt_lang], num_beams=num_beams, num_return_sequences=num_return_sequences
|
158 |
+
)
|
159 |
+
|
160 |
+
translations = [tokenizer.decode(translation, skip_special_tokens=True) for translation in translated_tokens]
|
161 |
+
return text, translations
|
162 |
+
|
163 |
# Test the translation
|
164 |
text = "Текст для перевода"
|
165 |
print(translate(text))
|