You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When the local model weight and the lora weight are saved in different locations, the code will try hf_hub_download and throw HFValidationError: Repo id must be in the form 'repo_name' error. Line 218 will throw this Error again, which leads the process to stop.
The text was updated successfully, but these errors were encountered:
Opdoop
changed the title
[BUG] Load local peft model while base model weight and lora weight save in different location
[BUG] Load local peft model while base model weight and lora weight are saved in different location
May 5, 2023
Opdoop
changed the title
[BUG] Load local peft model while base model weight and lora weight are saved in different location
[BUG] Load local peft model while base model weight and lora weight are saved in different locations
May 5, 2023
Opdoop
added a commit
to Opdoop/trl
that referenced
this issue
May 5, 2023
* Fix bug when loading local peft model
Fix bug in #341
* Fix loading bug when load lora mode
Fix loading bug when load lora model but not resuming training
1. Implement the fix logic described in #342 (review)
2. Set peft lora weight to trainable.
* Remove is_trainable
Leave is_trainable to future PR.
* add test_load_pretrained_peft
Check that the model saved with peft class interface can be loaded properly.
https://github.com/lvwerra/trl/blob/c85cdbdbd0d956bbc6f5a8e04b0036648480e7a1/trl/models/modeling_base.py#L205-L220
When the local model weight and the lora weight are saved in different locations, the code will try
hf_hub_download
and throwHFValidationError: Repo id must be in the form 'repo_name'
error. Line 218 will throw this Error again, which leads the process to stop.The text was updated successfully, but these errors were encountered: