fail to run model when load low bits instead of load original for qwen

[aoke79]

the error message is like below:
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, xpu:0 and cpu!
Thanks a lot,

[Oscilloscope98]

Hi @aoke79 ,

In your code, when using load_low_bit, is cpu_embedding set to True? If so, this may due to a bug with cpu_embedding. We have fixed it for bigdl-llm >= 2.5.0b20240123. Would you mind letting us know your bigdl-llm version? If bigdl-llm is below 2.5.0b20240123, you could update to latest bigdl-llm and have a try again :)

Please let us know for any further updates.