Skip to content

Don't throw error if ForwardContext is not available#730

Merged
calpt merged 3 commits intoadapter-hub:mainfrom
calpt:fix/quantized_train
Aug 10, 2024
Merged

Don't throw error if ForwardContext is not available#730
calpt merged 3 commits intoadapter-hub:mainfrom
calpt:fix/quantized_train

Conversation

@calpt
Copy link
Member

@calpt calpt commented Aug 7, 2024

Fixes in this PR:

  • Fix 'NoneType' error if context not available
  • Fix Llama grad checkpointing training.
  • Add quantized training test.

Still existing limitation:

  • ForwardContext does not work with gradient checkpointing, ie methods such as reft or prefix tuning don't work with gradient checkpointing currently.

Fix Llama grad checkpointing training.
Add quantized training test.
@calpt calpt changed the title Fix 'NoneType' error if context not available. Don't throw error if ForwardContext is not available Aug 7, 2024
@calpt calpt marked this pull request as ready for review August 8, 2024 07:59
Copy link
Member

@lenglaender lenglaender left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good to me! One last thing: The newly added test is currently skipped because in our pipeline we don't yet install bitsandbytes:

image

As soon as you have added this to the pipeline & the tests run through, you can merge it

@calpt calpt merged commit 251a853 into adapter-hub:main Aug 10, 2024
@calpt calpt deleted the fix/quantized_train branch August 10, 2024 12:04
dainis-boumber added a commit to ReDASers/adapters that referenced this pull request Aug 30, 2024
Fixes in this PR:
- Fix 'NoneType' error if context not available
- Fix Llama grad checkpointing training.
- Add quantized training test.

Still existing limitation:
- ForwardContext does not work with gradient checkpointing, ie methods
such as reft or prefix tuning don't work with gradient checkpointing
currently.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants