Update main.py

main.py

@@ -4,43 +4,35 @@ import pandas as pd
 from fastapi import FastAPI
 from pydantic import BaseModel
 from sentence_transformers import SentenceTransformer, util
-from langdetect import detect
-from huggingface_hub import hf_hub_download
+from huggingface_hub import snapshot_download
 
-# --- Cache configuration ---
+# --- Cache Configuration ---
 os.environ["HF_HOME"] = "/app/hf_cache"
 os.environ["TRANSFORMERS_CACHE"] = "/app/hf_cache"
-os.environ["TORCH_DISABLE_CUDA"] = "1"  # CPU only
+os.environ["TORCH_DISABLE_CUDA"] = "1"
 
-# --- Paths ---
-MODEL_PATH = './muril_combined_multilingual_model'
-CSV_PATH = './muril_multilingual_dataset.csv'
-HF_REPO = "Sp2503/muril-dataset"
-HF_FILE = "answer_embeddings.pt"
+# --- Hugging Face Repo ---
+HF_REPO = "Sp2503/Muril-Model"
 
-print("⚙️ Loading model and embeddings...")
+# --- Download model & embeddings from Hugging Face Hub ---
+print("📦 Downloading model & embeddings from Hugging Face Hub...")
+model_dir = snapshot_download(repo_id=HF_REPO, repo_type="model")
+print(f"✅ Model snapshot available at: {model_dir}")
 
-# Load MuRIL model
-model = SentenceTransformer(MODEL_PATH)
+MODEL_PATH = model_dir
+CSV_PATH = os.path.join(model_dir, "muril_multilingual_dataset.csv")
+EMBED_PATH = os.path.join(model_dir, "answer_embeddings.pt")
 
-# Load CSV dataset
+# --- Load resources ---
+print("⚙️ Loading model and embeddings...")
+model = SentenceTransformer(MODEL_PATH)
 df = pd.read_csv(CSV_PATH).dropna(subset=['question', 'answer'])
-
-# Load precomputed embeddings from Hugging Face
-hf_path = hf_hub_download(
-    repo_id=HF_REPO,
-    filename=HF_FILE,
-    repo_type="dataset",
-    cache_dir="/tmp"
-)
-answer_embeddings = torch.load(hf_path, map_location="cpu")
-
+answer_embeddings = torch.load(EMBED_PATH, map_location="cpu")
 print("✅ Model and embeddings loaded successfully.")
 
-# --- FastAPI app ---
+# --- FastAPI Setup ---
 app = FastAPI(title="MuRIL Multilingual QA API")
 
-# --- Request/Response models ---
 class QueryRequest(BaseModel):
     question: str
     lang: str = None
@@ -48,16 +40,14 @@ class QueryRequest(BaseModel):
 class QAResponse(BaseModel):
     answer: str
 
-# --- Root endpoint ---
 @app.get("/")
 def root():
-    return {"status": "✅ Running MuRIL QA API", "model_loaded": True}
+    return {"status": "✅ API ready", "model_loaded": True}
 
-# --- QA endpoint ---
 @app.post("/get-answer", response_model=QAResponse)
 def get_answer_endpoint(request: QueryRequest):
     question_text = request.question.strip()
-    lang_filter = request.lang or detect(question_text)
+    lang_filter = request.lang
 
     filtered_df = df
     filtered_embeddings = answer_embeddings
@@ -67,7 +57,7 @@ def get_answer_endpoint(request: QueryRequest):
         filtered_embeddings = answer_embeddings[mask.values]
 
     if len(filtered_df) == 0:
-        return {"answer": f"⚠️ No data found for language '{lang_filter}'."}
+        return {"answer": f"No data found for language '{lang_filter}'."}
 
     question_emb = model.encode(question_text, convert_to_tensor=True)
     cosine_scores = util.pytorch_cos_sim(question_emb, filtered_embeddings)
@@ -75,7 +65,6 @@ def get_answer_endpoint(request: QueryRequest):
     answer = filtered_df.iloc[best_idx]['answer']
     return {"answer": answer}
 
-# --- Run app ---
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run("main:app", host="0.0.0.0", port=8080)