[bug] of convert llama-30b using lmquant.llm.run

[lisuying214]

when I want to convert llama-30b model, I came across the bug
File "/xxx/deepcompressor/lmquant/quant/calib/rotate.py", line 152, in transform_rms_norm_and_linear
linear.weight.data = (fc_w * ln_w).to(dtype=dtype)
RuntimeError: Attempted to call variable.set_data(tensor), but variable and tensor have incompatible tensor type.

I found that the fc_w and ln_w device.type is "meta", leading to this error. Do you have similar problem? Hope for your reply and suggestion.

ps: the model is downloaded by "git clone https://gitee.com/hf-models/llama-30b"