Changes to the trainer to minimise potential mismatch with vllm #1167
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
samsja
reviewed
Oct 27, 2025
samsja
reviewed
Oct 27, 2025
samsja
reviewed
Oct 27, 2025
Co-authored-by: samsja <[email protected]> Signed-off-by: faresobeid <[email protected]>
Co-authored-by: samsja <[email protected]> Signed-off-by: faresobeid <[email protected]>
Signed-off-by: faresobeid <[email protected]>
Removed unused import of Optional from typing. Signed-off-by: faresobeid <[email protected]>
Signed-off-by: faresobeid <[email protected]>
Signed-off-by: faresobeid <[email protected]>
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Outdated suggestions cannot be applied.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.
Currently have matched rope implementations (seemingly vllm does rope in bf16 instead of fp32 which is weird)
Add residuals in fp32 instead of bf16
Stop using liger kernel for our RMS Norm
Planned:
Check MoE difference
Vllm fuses the residual and rms norm, not sure if we can do the equivalent
torch.topk apparently is unstable so will try to find an alternative
Test actual mismatch_kl difference and mfu difference (although mismatch_kl here might not be perfect as it can have an impact late in training)