Fix: update from disable_adapter to disable_adapters #221
Conversation
|
Hi we'd like to be able to reproduce the issue on our side in order to evaluate the PR. Will also comment on the issue you opened, unslothai/unsloth#3067, for further discussion. |
| logits_to_keep = logits_to_keep + 1, | ||
| ).logits | ||
| with torch.inference_mode(): | ||
| unwarp_model = trainer.accelerator.unwrap_model(trainer.model, keep_fp32_wrapper = False) |
There was a problem hiding this comment.
instead of an if else, could we create a conditional context? Something like
from contextlib import nullcontext
unwrap_model = trainer.accelerator.unwrap_model(trainer.model, keep_fp32_wrapper=False)
try:
disable_adapter_ctx = base.disable_adapter()
except AttributeError:
disable_adapter_ctx = nullcontext()
There was a problem hiding this comment.
Thanks for your comments!
- I'm sorry, what is
base? - I think
try exceptshould be the last method to process this issue, I'm not sure if it's a good time to use it right now.
| ).logits | ||
| with torch.inference_mode(): | ||
| unwarp_model = trainer.accelerator.unwrap_model(trainer.model, keep_fp32_wrapper = False) | ||
| if unwarp_model._hf_peft_config_loaded: |
There was a problem hiding this comment.
just a small nit: unwarp_model should be unwrap_model
|
Hi @wa008, I made some code comments. But one issue is that it should not be calling If you could update based on the comments I made we should be good to go. |
Thanks for your reply and suggestions. I’m sorry, I didn’t fully understand what you meant. If you’d like, you can go ahead and submit a new PR for this issue—I’m completely fine with that. |
Solve unlsoth issue-3064
relate link:
disable_adaptersfunction