Upload server.py with huggingface_hub

server.py

@@ -21,10 +21,9 @@ def load_model():
     pipe = GlmImagePipeline.from_pretrained(
         "zai-org/GLM-Image",
         torch_dtype=torch.bfloat16,
-        device_map="cuda",
-        enable_model_cpu_offload=True,
     )
-    print("[image] Model loaded successfully!", flush=True)
+    pipe.enable_sequential_cpu_offload()
+    print("[image] Model loaded with sequential CPU offload!", flush=True)
 
 @app.route("/health", methods=["GET"])
 def health():