Remove forward_backward batch cap from 256k Qwen3.8-27B definition - #64
Merged
Merged
Conversation
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Contributor
|
I'll fix CI failures and address comments from users with write access that start with 'DevinAI' or '@devin'.
|
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Drops
MAX_FORWARD_BACKWARD_BATCH = 1fromqwen3_8_27b_miles_lora_256k.pyso the engine uses its default (uncapped) coalescing: a step's 128 SDK chunks become oneforward_backwardop with batch 128 instead of 128 single-datum ops.The cap was added in #39 as a guard against a suspected process-group wedge from long collective chains across both nodes. A 3-step probe on the matched 2-node TP2×CP8 topology with the cap removed ran clean (
forward_backward_calls= 1/step,mean_batch= 128, no NCCL timeout/wedge/OOM), so the guard isn't needed. Performance is neutral: trainer-sideforward_backward_s≈ 2256 s in both configs; client train time 2470 s vs 2404 s capped over a 3-step sample.Probe: https://wandb.ai/modal-labs/miles-lora-longcontext/runs/lilo-27b-256k-2n-nocap-3step-1790117905-1790117910
Link to Devin session: https://modal.devinenterprise.com/sessions/d017b320e7e1430aadae6c51e49c3029
Open in Devin Desktop: https://modal.devinenterprise.com/desktop/session/d017b320e7e1430aadae6c51e49c3029?variant=devin
Requested by: @micahtyong