Update handler.py

handler.py

@@ -360,6 +360,11 @@ def generate_response(
 def query(payload: dict):
     """HF Endpoint entry (demo-like)."""
     global model_initialized, tokenizer, model, image_processor, context_len, args
+
+    # 🟢 Health check kısayolu: hem {"health_check": true} hem de {"message": "health_check"} desteklenir
+    if payload.get("health_check") or payload.get("message") == "health_check":
+        return health_check()
+        
     if not model_initialized:
         if not initialize_model():
             return {"error": "Model initialization failed"}
@@ -396,14 +401,38 @@ def query(payload: dict):
         return {"error": f"Query failed: {e}"}
 
 def health_check():
-    return {
+    info = {
         "status": "healthy",
         "model_initialized": model_initialized,
-        "cuda_available": torch.cuda.is_available(),
         "llava_available": LLAVA_AVAILABLE,
         "transformers_available": TRANSFORMERS_AVAILABLE,
+        "cuda_available": torch.cuda.is_available(),
     }
 
+    if torch.cuda.is_available():
+        try:
+            device_index = torch.cuda.current_device()
+            props = torch.cuda.get_device_properties(device_index)
+            total_vram_gb = round(props.total_memory / (1024 ** 3), 2)
+            used_vram_gb = round(torch.cuda.memory_allocated(device_index) / (1024 ** 3), 2)
+            reserved_vram_gb = round(torch.cuda.memory_reserved(device_index) / (1024 ** 3), 2)
+
+            info.update({
+                "cuda_device_index": device_index,
+                "cuda_name": props.name,
+                "cuda_compute_capability": f"{props.major}.{props.minor}",
+                "cuda_total_vram_gb": total_vram_gb,
+                "cuda_used_vram_gb": used_vram_gb,
+                "cuda_reserved_vram_gb": reserved_vram_gb,
+                "torch_version": torch.__version__,
+                "cuda_runtime_version": torch.version.cuda,
+            })
+        except Exception as e:
+            info["cuda_error"] = str(e)
+
+    return info
+
+
 def get_model_info():
     if not model_initialized:
         return {"error": "Model not initialized"}