Allow for preventing 16-bit cast of marked modules #61
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
Quentin-Anthony
approved these changes
Mar 8, 2024
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Outdated suggestions cannot be applied.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.
This PR allows for one to mark a given model parameter with
p._deepspeed_no_cast = Trueto allow it to not be cast into 16-bit by DeepSpeed.Used this to train a mamba-160m, keeping A_log and D params in fp32 throughout, which performed on-par with paper's reported results. However, if this change seems like a bad idea or is likely to break things, would be glad to know what to check for.
This should also allow us to keep
inv_freqin rotary embeddings in fp32!