Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Appearance settings

Comments

Close side panel

Ring Attention First Pass Implementation#407

Open
pulitha13 wants to merge 1 commit intofoundation-model-stack:mainfoundation-model-stack/foundation-model-stack:mainfrom
coms-6998-context-parallelism-project:fms-pr-squashedcoms-6998-context-parallelism-project/foundation-model-stack:fms-pr-squashedCopy head branch name to clipboard
Open

Ring Attention First Pass Implementation#407
pulitha13 wants to merge 1 commit intofoundation-model-stack:mainfoundation-model-stack/foundation-model-stack:mainfrom
coms-6998-context-parallelism-project:fms-pr-squashedcoms-6998-context-parallelism-project/foundation-model-stack:fms-pr-squashedCopy head branch name to clipboard

Conversation

@pulitha13
Copy link

This is a first pass in using context parallelism for FMS based on the ring attention paper here. We are treating this pull request as a work in progress. We would be happy to make any changes or fixes as IBM sees fit

Known issues

  • We are not perfectly generating 1:1 output logits when compared to the default configuration of llama 7b. We aim to figure out why our results are deviating yet producing "close enough" output.

Future improvements:

  • Online softmaxing during ring attention

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.

1 participant

Morty Proxy This is a proxified and sanitized view of the page, visit original site.