Add model with Git LFS

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+model/** filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,87 @@
+from fastapi import FastAPI, Request
+from pydantic import BaseModel
+from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
+import torch
+import re
+
+app = FastAPI()
+
+def split_by_words_and_dot(text, min_words=125, max_words=160, fallback_words=150):
+    import re
+    words = re.findall(r'\S+|\n', text)  # giữ nguyên \n như một "từ"
+    chunks = []
+    start = 0
+    while start < len(words):
+        end = min(start + max_words, len(words))
+        # Tìm dấu chấm trong khoảng min_words đến max_words
+        dot_idx = -1
+        for i in range(start + min_words, min(start + max_words, len(words))):
+            if words[i] == '.' or (words[i].endswith('.') and words[i] != '\n'):
+                dot_idx = i
+        if dot_idx != -1:
+            chunk_end = dot_idx + 1
+        elif end - start > fallback_words:
+            chunk_end = start + fallback_words
+        else:
+            chunk_end = end
+        chunk = ' '.join([w if w != '\n' else '\n' for w in words[start:chunk_end]]).replace(' \n ', '\n').replace(' \n', '\n').replace('\n ', '\n')
+        chunks.append(chunk.strip())
+        start = chunk_end
+    return chunks
+
+# Load model
+model_path = "./model/facebook-m2m100_418M-fine_tuning"
+tokenizer = M2M100Tokenizer.from_pretrained(model_path)
+model = M2M100ForConditionalGeneration.from_pretrained(model_path)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
+
+class TranslateRequest(BaseModel):
+    text: str
+    source_lang: str
+    target_lang: str
+
+# @app.post("/translate")
+# def translate_text(req: TranslateRequest):
+#     tokenizer.src_lang = req.source_lang
+#     encoded = tokenizer(req.text, return_tensors="pt", truncation=True, max_length=512).to(device)
+#     generated_tokens = model.generate(
+#         **encoded,
+#         forced_bos_token_id=tokenizer.get_lang_id(req.target_lang),
+#         max_length=512,      # tăng lên nếu cần dịch đoạn dài, nhưng không nên quá lớn
+#         num_beams=2,         # giảm beam search để nhanh hơn
+#         no_repeat_ngram_size=3,
+#         early_stopping=True
+#     )
+#     translated_text = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
+#     return {
+#         "source_text": req.text,
+#         "translated_text": translated_text,
+#         "src_lang": req.source_lang,
+#         "tgt_lang": req.target_lang
+#     }
+
+
+@app.post("/translate")
+def translate_text(req: TranslateRequest):
+    tokenizer.src_lang = req.source_lang
+    text_chunks = split_by_words_and_dot(req.text, min_words=125, max_words=160, fallback_words=150)
+    translated_chunks = []
+    for chunk in text_chunks:
+        encoded = tokenizer(chunk, return_tensors="pt", truncation=True, max_length=256).to(device)
+        generated_tokens = model.generate(
+            **encoded,
+            forced_bos_token_id=tokenizer.get_lang_id(req.target_lang),
+            max_length=256,
+            num_beams=2,
+            no_repeat_ngram_size=3,
+        )
+        translated_text = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
+        translated_chunks.append(translated_text)
+    full_translation = "\n".join(translated_chunks)
+    return {
+        "source_text": req.text,
+        "translated_text": full_translation,
+        "src_lang": req.source_lang,
+        "tgt_lang": req.target_lang
+    }

dockerfile

@@ -0,0 +1,14 @@
+FROM python:3.10-slim-bullseye
+
+WORKDIR /app
+
+# Cài g++ để tránh lỗi transformers build lại tokenizer
+RUN apt-get update && apt-get install -y g++ && apt-get upgrade -y && apt-get clean && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+# Mặc định FastAPI chạy bằng Uvicorn
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

requirements.txt

@@ -0,0 +1,4 @@
+fastapi
+uvicorn
+torch
+transformers