Update app.py

app.py

@@ -1,32 +1,53 @@
+import os
 from fastapi import FastAPI
-from pydantic import BaseModel
+import uvicorn
 from sentence_transformers import SentenceTransformer
 from threading import Thread
+from typing import Optional
 
-app = FastAPI()
+# Initialize FastAPI
+app = FastAPI(title="Sentence Transformer API")
+
+# Global model variable
+model: Optional[SentenceTransformer] = None
 
-# Load model in background (avoids startup timeout)
-model = None
 def load_model():
+    """Background thread for model loading"""
     global model
-    model = SentenceTransformer("BAAI/bge-small-zh")
+    try:
+        model = SentenceTransformer('all-MiniLM-L6-v2')
+        print("Model loaded successfully")
+    except Exception as e:
+        print(f"Model loading failed: {str(e)}")
 
-Thread(target=load_model).start()
+# Start model loading immediately
+Thread(target=load_model, daemon=True).start()
 
-class Input(BaseModel):
-    texts: list[str]
+@app.get("/")
+def health_check():
+    """Required health check endpoint"""
+    return {
+        "status": "ready",
+        "model_loaded": model is not None
+    }
 
 @app.post("/embed")
-def embed(input: Input):
+async def embed_text(text: str):
+    """Endpoint for text embeddings"""
     if model is None:
-        return {"error": "Model not loaded yet"}, 503
-    vectors = model.encode(input.texts, normalize_embeddings=True).tolist()
-    return {"embeddings": vectors}
-
-@app.get("/")
-def health():
-    return {"status": "OK"}  # Critical for health checks
+        return {"error": "Model still loading"}, 503
+    embeddings = model.encode(text)
+    return {
+        "text": text,
+        "embedding": embeddings.tolist(),
+        "dimension": len(embeddings)
+    }
 
 if __name__ == "__main__":
-    import uvicorn
-    uvicorn.run(app, host="0.0.0.0", port=8080)
+    port = int(os.environ.get("PORT", 7860))  # Must be 7860 for Spaces
+    uvicorn.run(
+        "app:app",
+        host="0.0.0.0",
+        port=port,
+        reload=False  # Disable auto-reload in production
+    )