update: minors corrections

.dockerignore

@@ -0,0 +1 @@
+venv

Dockerfile

@@ -1,14 +1,21 @@
 FROM python:3.10-slim
-# FROM pytorch/pytorch:2.1.0-cuda11.8-cudnn8-runtime
+# FROM pytorch/pytorch:2.4.1-cuda12.4-cudnn9-runtime
 
 WORKDIR /app
 
 RUN apt-get update && apt-get install -y git ffmpeg
+# RUN sed -i 's|http://archive.ubuntu.com/ubuntu|http://fr.archive.ubuntu.com/ubuntu|g' /etc/apt/sources.list && \
+#     apt-get update && apt-get install -y --no-install-recommends git && \
+#     rm -rf /var/lib/apt/lists/*
 
 COPY . .
 
+RUN pip install --no-cache-dir --upgrade pip
 RUN pip install --no-cache-dir -r requirements.txt
 
-# EXPOSE 8000
-# CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
-CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]
+# CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]
+
+# docker buildx build --platform linux/amd64 --no-cache -t ai-api .
+# docker tag ai-api kora3/ai-api:latest
+# docker push kora3/ai-api:latest

functions/translation.py

@@ -1,23 +1,25 @@
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import torch
 
-MODEL_NAME = "facebook/nllb-200-distilled-600M"
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
+MODEL_NAME = "facebook/nllb-200-distilled-600M" # facebook/nllb-200-3.3B
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME).to(device)
 model.eval()
 
 
 def translateText(text: str, sourceLang: str, targetLang: str) -> str:
     tokenizer.src_lang = sourceLang
-    inputs = tokenizer(text, return_tensors="pt")
+    inputs = tokenizer(text, return_tensors="pt").to(device)
 
     with torch.no_grad():
         tokens = model.generate(
             **inputs,
             forced_bos_token_id=tokenizer.convert_tokens_to_ids(targetLang),
-            num_beams=1,
-            max_length=128
+            max_new_tokens=512,
+            num_beams=4,
+            early_stopping=True,
         )
 
     translatedText = tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]

language/dioula/dyu_stt.py

@@ -1,9 +1,11 @@
-import base64, tempfile, os
+import base64, tempfile, os, torch
 from transformers import pipeline
 from functions.utils import getAudioDuration
 
 MODEL_NAME = "facebook/mms-1b-all"
-pipe = pipeline("automatic-speech-recognition", model=MODEL_NAME, model_kwargs={"target_lang": "dyu"})
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+pipe = pipeline("automatic-speech-recognition", model=MODEL_NAME, model_kwargs={"target_lang": "dyu"}, device=device)
 
 
 def dioulaSTT(audioBase64: str) -> dict:
@@ -20,4 +22,4 @@ def dioulaSTT(audioBase64: str) -> dict:
     finally:
         os.remove(tempAudioPath)
 
-    return {'text': text, 'language': 'mos', 'duration': duration}
+    return {'text': text, 'language': 'dyu', 'duration': duration}

language/fr_mos.py

@@ -0,0 +1,23 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+
+MODEL_NAME = "code-li/nllb-moore"  
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME).to(device)
+model.eval()
+
+
+def translateFRMOS(text: str, sourceLang: str, targetLang: str):
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, src_lang=sourceLang)
+    TGT_LANG_ID = tokenizer.convert_tokens_to_ids(targetLang)
+
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=128).to(device)
+    with torch.no_grad():
+        tokens = model.generate(
+            **inputs,
+            forced_bos_token_id=TGT_LANG_ID,
+            max_length=128,
+            num_beams=4,
+        )
+    return tokenizer.decode(tokens[0], skip_special_tokens=True)

language/moore/mos_stt.py

@@ -1,11 +1,12 @@
-import base64, tempfile, os
+import base64, tempfile, os, torch
 from transformers import pipeline
 from functions.utils import getAudioDuration
-# from huggingface_hub import login
 
 MODEL_NAME = "facebook/mms-1b-all"
-pipe = pipeline("automatic-speech-recognition", model=MODEL_NAME, model_kwargs={"target_lang": "mos"})
-# login(token=os.environ["HF_TOKEN"])
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+pipe = pipeline("automatic-speech-recognition", model=MODEL_NAME, model_kwargs={"target_lang": "mos"}, device=device)
+
 # MODEL_NAME = "burkimbia/BIA-WHISPER-LARGE-SACHI_V3"
 # pipe = pipeline("automatic-speech-recognition", model=MODEL_NAME)
 

main.py

@@ -2,11 +2,19 @@ from fastapi import FastAPI, Request, HTTPException
 from functions.translation import translateText
 from functions.speech_to_text import speechToText
 from functions.text_to_speech import textToSpeech
+
 from language.moore.mos_stt import mooreSTT
 from language.moore.mos_tts import mooreTTS
+
 from language.dioula.dyu_stt import dioulaSTT
 from language.dioula.dyu_tts import dioulaTTS
 
+from language.fr_mos import translateFRMOS
+
+import os
+from huggingface_hub import login
+login(token=os.environ["HF_TOKEN"])
+
 
 
 app = FastAPI(
@@ -30,6 +38,21 @@ async def translate(request: Request):
         raise HTTPException(status_code=400, detail=f"Translate error: {e}")
 
 
+@app.post("/codeLi/translateText")
+async def translateMoore(request: Request):
+    body: dict = await request.json()
+    try:
+        text = body.get('text')
+        sourceLang = body.get('sourceLang')
+        targetLang = body.get('targetLang')
+
+        translatedText = translateFRMOS(text=text, sourceLang=sourceLang, targetLang=targetLang)
+        return { 'translatedText': translatedText }
+    except Exception as e:
+        print(f"Translate mos error: {e}")
+        raise HTTPException(status_code=400, detail=f"Translate mos error: {e}")
+
+
 @app.post("/whisper/speechToText")
 async def whisperSpeechToText(request: Request):
     body: dict = await request.json()

requirements.txt

@@ -86,7 +86,7 @@ networkx==2.8.8
 nltk==3.9.4
 num2words==0.5.14
 numba==0.65.1
-numpy==1.22.0
+numpy>=1.23.2,<2.0.0
 onnxruntime==1.23.2
 packaging==26.2
 pandas==1.5.3