[New Feature] finetune local HuggingFace model

Open dwang888 opened this issue 2 years ago • 1 comments

In some circumstance, Huggingface is blocked from being downloaded directly, e.g. firewall. Is it feasible to add an option to load local HuggingFace model rather than downloading from hub directly? In "examples/finetune.py", the model is loaded as "model = AutoModel.get_model(model_args)", if it also support local path OR model name, it will be highly useful. Thanks in advance!

Jun 20 '23 21:06 dwang888

Hi, LMFlow and Huggingface Transformers support finetuning local models natively. You can directly replace the model path with your local path. Hope this helps. Thank you!

Jun 22 '23 15:06 shizhediao

This issue has been marked as stale because it has not had recent activity. If you think this still needs to be addressed please feel free to reopen this issue. Thanks

Sep 30 '23 19:09 shizhediao

thank you!

Sep 30 '23 19:09 dwang888