update

Dockerfile

@@ -1,17 +1,23 @@
-# Use an official Python runtime as a parent image
-FROM python:3.9-slim
+FROM python:3.10-slim
 
-# Set the working directory in the container
-WORKDIR /code
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
 
-# Copy the dependencies file to the working directory
-COPY requirements.txt .
+WORKDIR /app
 
-# Install any needed packages specified in requirements.txt
-RUN pip install --no-cache-dir --upgrade -r requirements.txt
+# system deps for sentencepiece & build
+RUN apt-get update && apt-get install -y git build-essential
 
-# Copy the rest of the application code to the working directory
-COPY . .
+# copy code
+COPY . /app
 
-# Command to run the API
-CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
+# install python deps
+RUN pip install --upgrade pip
+RUN pip install -r requirements.txt
+
+# Ensure model is downloaded at build-time to speed cold-start (optional)
+# You can uncomment the following to pre-cache HF model in the image:
+RUN python -c "from transformers import AutoTokenizer, AutoModelForSeq2SeqLM; AutoTokenizer.from_pretrained('Helsinki-NLP/opus-mt-en-ar'); AutoModelForSeq2SeqLM.from_pretrained('Helsinki-NLP/opus-mt-en-ar')"
+
+# Use gunicorn for production inside Space
+CMD ["gunicorn", "--bind", "0.0.0.0:7860", "app:app", "--workers", "1", "--threads", "4", "--timeout", "300"]

app.py

@@ -1,46 +1,64 @@
-from fastapi import FastAPI, HTTPException
-from pydantic import BaseModel
-import ctranslate2
-import sentencepiece as spm
+# app.py
+from flask import Flask, request, jsonify
+from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
 import os
 
-app = FastAPI()
+app = Flask(__name__)
 
-# Load the CTranslate2 model and the SentencePiece tokenizer
-try:
-    model_path = "en_ar_ct2_model/"
-    sp_model_path = os.path.join(model_path, "source.spm")
+MODEL_ID = os.environ.get("MODEL_ID", "Helsinki-NLP/opus-mt-en-ar")
 
-    translator = ctranslate2.Translator(model_path, device="cpu") # Use "cuda" if on a GPU Space
-    sp = spm.SentencePieceProcessor()
-    sp.load(sp_model_path)
-except Exception as e:
-    # This helps in debugging if the model files are not found
-    raise RuntimeError(f"Error loading model: {e}")
+# Lazy load on first request (avoid heavy imports on cold boot if you prefer)
+translator = None
 
+def get_translator():
+    global translator
+    if translator is None:
+        # Load tokenizer + model explicitly to control device/kwargs if needed
+        tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+        model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)
+        translator = pipeline("translation", model=model, tokenizer=tokenizer, src="en", tgt="ar")
+    return translator
 
-class TranslationRequest(BaseModel):
-    text: list[str] # Expect a list of strings for batching
+@app.route("/health", methods=["GET"])
+def health():
+    return jsonify({"status": "ok"}), 200
 
-class TranslationResponse(BaseModel):
-    translations: list[str]
+@app.route("/translate", methods=["POST"])
+def translate():
+    """
+    Accepts JSON:
+    {
+      "texts": ["Hello", "How are you?"],   # or a single string as "text"
+      "max_length": 256,                    # optional
+      "batch_size": 8                       # optional
+    }
+    Returns:
+    {
+      "translations": ["مرحبا", "كيف حالك؟"]
+    }
+    """
+    payload = request.get_json(force=True)
+    if payload is None:
+        return jsonify({"error": "invalid json"}), 400
 
-@app.get("/")
-def read_root():
-    return {"message": "English to Arabic Translation API is running."}
+    # allow single text or list
+    texts = payload.get("texts") or payload.get("text")
+    if texts is None:
+        return jsonify({"error": "provide 'text' or 'texts' in JSON"}), 400
 
-@app.post("/translate", response_model=TranslationResponse)
-def translate_text(request: TranslationRequest):
-    if not request.text:
-        raise HTTPException(status_code=400, detail="Input text list cannot be empty.")
+    if isinstance(texts, str):
+        texts = [texts]
 
-    source_sentences = request.text
-    source_tokenized = [sp.encode(sentence, out_type=str) for sentence in source_sentences]
+    max_length = payload.get("max_length", 256)
+    batch_size = payload.get("batch_size", 8)
 
-    # Translate the batch of sentences
-    translations_tokenized = translator.translate_batch(source_tokenized)
+    pipe = get_translator()
+    # pipeline supports batched translation
+    translated = pipe(texts, max_length=max_length, batch_size=batch_size)
+    # pipeline returns list of dicts like {"translation_text": "..."}
+    out = [t["translation_text"] for t in translated]
+    return jsonify({"translations": out}), 200
 
-    # Decode the translated sentences
-    translations = [sp.decode(translation.hypotheses[0]) for translation in translations_tokenized]
-
-    return {"translations": translations}
+if __name__ == "__main__":
+    # For local debug only; Spaces uses gunicorn (Dockerfile will define)
+    app.run(host="0.0.0.0", port=int(os.environ.get("PORT", 8080)))

convert_model.py

@@ -0,0 +1,24 @@
+import os
+from ct2_transformers import TransformerConverter
+
+# 1. Define the source Hugging Face model
+model_name = "Helsinki-NLP/opus-mt-tc-big-en-ar"
+
+# 2. Define the output path for the converted model
+output_dir = "en_ar_ct2_model"
+
+# Create the output directory if it doesn't exist
+if not os.path.exists(output_dir):
+    os.makedirs(output_dir)
+
+print(f"Starting conversion of model: {model_name}")
+print("This may take a few moments...")
+
+# 3. Initialize the converter
+converter = TransformerConverter(model_name)
+
+# 4. Run the conversion and apply quantization for speed
+converter.convert(output_dir, quantization="int8")
+
+print(f"\nModel successfully converted and saved to the '{output_dir}' folder.")
+print("You are now ready to upload this folder to your Hugging Face Space.")

requirements.txt

@@ -1,5 +1,5 @@
-fastapi
-uvicorn
-ctranslate2
+flask
+transformers>=4.30
+torch              # or torch-cpu depending on environment
 sentencepiece
-transformers
+gunicorn