Tweak Pipeline Parallel layer split strategy #20
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
log "No checkpoint path provided" only on rank 0
remove torch compile
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The previous pipelineblock split strategy in nanotron tries to evenly split the layers to each device in terms of FLOPs. However, since the embedding table is becoming larger and larger, memory consumption should be taken into account.
We follow the practice in Llama 405B training, which treat the embedding table and lm head as heavy as a single transformer block. (This is why 405B model has 126 layers.)
Any suggestions for better general strategy?
e.g. a 32-layer llama model and PP=4:
Prev:
After:
