Skip to content

PaddingFree#113

Closed
RhuiDih wants to merge 7 commits intoinstructlab:mainfrom
RhuiDih:dev/model_mod
Closed

PaddingFree#113
RhuiDih wants to merge 7 commits intoinstructlab:mainfrom
RhuiDih:dev/model_mod

Conversation

@RhuiDih
Copy link
Copy Markdown

@RhuiDih RhuiDih commented Jun 28, 2024

No description provided.

@RhuiDih RhuiDih force-pushed the dev/model_mod branch 3 times, most recently from dc9d768 to 2868eab Compare June 28, 2024 08:40
@fabianlim
Copy link
Copy Markdown
Collaborator

fabianlim commented Jul 1, 2024

Performance testing:

Legend:

  • ilab-granite: using dolomite model in ilab training script
  • ilab-llama: using huggingface model in illab training script
  • hf-llama: using the padding free introduced in this PR

image

After removing dropout from Dolomite, we can achive the same MT bench score

image

fabianlim and others added 2 commits July 2, 2024 08:08
Signed-off-by: Yu Chin Fabian Lim <flim@sg.ibm.com>
@fabianlim
Copy link
Copy Markdown
Collaborator

This may be replaced by the official HF solution once its merged huggingface/transformers#31446

@Maxusmusti Maxusmusti closed this Sep 27, 2024
@mergify mergify Bot added the ci-failure label Sep 27, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants