skip_init causes bugs when finetuning using deepspeed.

#37
by sunworshipper - opened

modeling_chatglm.py 中模型初始化时调用的 skip_init,会导致 deepspeed报错。

sunworshipper changed discussion title from skip_init caused bugs when finetuning using deepspeed. to skip_init causes bugs when finetuning using deepspeed.
Knowledge Engineering Group (KEG) & Data Mining at Tsinghua University org

You can pass empty_init=False to AutoModel.from_pretrained to disable skip_init

zxdu20 changed discussion status to closed

Sign up or log in to comment