Use device_map to force CUDA placement and prevent meta device

acestep/handler.py

@@ -498,8 +498,7 @@ class AceStepHandler:
                                 trust_remote_code=True,
                                 attn_implementation=candidate,
                                 torch_dtype=self.dtype,
-                                low_cpu_mem_usage=False,  # Disable memory-efficient weight loading
-                                _fast_init=False,  # Disable fast initialization
+                                device_map={"": self.device},  # Force all components to target device immediately
                             )
                             attn_implementation = candidate
                             break