Qi Dai
Results
2
comments of
Qi Dai
> [@qidai2000](https://github.com/qidai2000) This is a bug in deepspeed. Some extra parameters are stored in the checkpoint files. You can extract the lora state dict using the following code. > >...
And for fine-tune, the problem remains the same. The model type of the finetuned models fail to be detected and "No models are loaded".