[70B-Part3] FSDP Training #109
Merged
Conversation
This file contains hidden or 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
farzadab
commented
Sep 10, 2024
juberti
reviewed
Sep 12, 2024
|
@juberti are there any more comments? |
juberti
approved these changes
Sep 16, 2024
akshat0311
pushed a commit
to jiviai/audio-llm
that referenced
this pull request
Jan 30, 2025
* use_fsdp option * return move to(device) when not using FSDP
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
This PR is mostly complete, but I do need to revisit some minor assumptions to make sure I didn't break the normal (non-fsdp) pipeline.
One of the odd changes here is the removal of
.to(device, dtype).dtypepart is moved into the model itself since it'll cost us too much to possibly be sloppy and load a full-precision 70B and then turn it into half precision.devicepart is being handled by thetraineritself when doingtrainer.train. This causes some minor issues (e.g. runningtrainer.evaluatebefore is not allowed), but they're not hugely important.model.deviceiscpu(or mps) beforetrainer.trainand thencuda:rankaftertrainer.train. This might take some getting used to.