Update app.py

app.py

@@ -34,15 +34,19 @@ if not hf_token:
     raise HTTPException(status_code=500, detail="HF_TOKEN environment variable not set")
 
 # Initialize Hugging Face Inference Client with primary and fallback models
-primary_model = "Qwen/Qwen2-7B-Instruct"
-fallback_model = "mistral/Mixtral-8x7B-Instruct-v0.1"
+primary_model = "mistral/Mixtral-8x7B-Instruct-v0.1"
+fallback_model = "Qwen/Qwen2-7B-Instruct"
 try:
     client = InferenceClient(model=primary_model, token=hf_token, provider="auto")
     logger.info(f"Hugging Face Inference Client initialized for {primary_model} with provider='auto'")
 except Exception as e:
     logger.warning(f"Failed to initialize client for {primary_model}: {str(e)}. Falling back to {fallback_model}")
-    client = InferenceClient(model=fallback_model, token=hf_token, provider="auto")
-    logger.info(f"Hugging Face Inference Client initialized for {fallback_model} with provider='auto'")
+    try:
+        client = InferenceClient(model=fallback_model, token=hf_token, provider="hf-inference")
+        logger.info(f"Hugging Face Inference Client initialized for {fallback_model} with provider='hf-inference'")
+    except Exception as e:
+        logger.error(f"Failed to initialize client for {fallback_model}: {str(e)}")
+        raise HTTPException(status_code=500, detail=f"Failed to initialize Inference Client: {str(e)}")
 
 # In-memory caches (1-hour TTL)
 raw_text_cache = cachetools.TTLCache(maxsize=100, ttl=3600)
@@ -97,9 +101,9 @@ async def process_pdf_page(img, page_idx):
         return ""
 
 async def process_with_qwen(filename: str, raw_text: str):
-    """Process raw text with Qwen2-7B-Instruct or fallback via Hugging Face Inference API."""
+    """Process raw text with LLM via Hugging Face Inference API."""
     start_time = time.time()
-    logger.info(f"Starting Qwen API processing for {filename}, {log_memory_usage()}")
+    logger.info(f"Starting LLM API processing for {filename}, {log_memory_usage()}")
 
     # Check structured data cache
     text_hash = get_text_hash(raw_text)
@@ -144,11 +148,11 @@ async def process_with_qwen(filename: str, raw_text: str):
         json_str = llm_output[json_start:json_end]
         structured_data = json.loads(json_str)
         structured_data_cache[text_hash] = structured_data
-        logger.info(f"Qwen API processing for {filename}, took {time.time() - start_time:.2f} seconds, {log_memory_usage()}")
+        logger.info(f"LLM API processing for {filename}, took {time.time() - start_time:.2f} seconds, {log_memory_usage()}")
         return structured_data
     except Exception as e:
-        logger.error(f"Qwen API processing failed for {filename}: {str(e)}, {log_memory_usage()}")
-        return {"error": f"Qwen API processing failed: {str(e)}"}
+        logger.error(f"LLM API processing failed for {filename}: {str(e)}, {log_memory_usage()}")
+        return {"error": f"LLM API processing failed: {str(e)}"}
 
 @app.post("/ocr")
 async def extract_and_structure(files: List[UploadFile] = File(...)):
@@ -263,7 +267,7 @@ async def extract_and_structure(files: List[UploadFile] = File(...)):
             except Exception as e:
                 logger.warning(f"Text normalization failed for {file.filename}: {str(e)}, {log_memory_usage()}")
 
-        # Process with Qwen API
+        # Process with LLM API
         structured_data = await process_with_qwen(file.filename, raw_text)
         success_count += 1
         output_json["data"].append({