[NPU] Remove enforce-eager for Qwen3.5-35B-A3B - #417
Merged
Merged
Conversation
Signed-off-by: gaofulin <gaofulin1@huawei.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request updates the VLLM arguments in the script scripts/run-qwen3.5-35B-A3B-npu.sh by increasing the GPU memory utilization from 0.7 to 0.85 and removing the --vllm-enforce-eager flag. There are no review comments, so I have no feedback to provide.
Documentation build overview
48 files changed ·
|
Fulin-Gao
marked this pull request as ready for review
September 10, 2026 08:31
Signed-off-by: gaofulin <gaofulin1@huawei.com>
CalvinXKY
approved these changes
Sep 14, 2026
Meihan-chen
added a commit
to Meihan-chen/vime
that referenced
this pull request
Sep 15, 2026
Merge upstream/ascend through 8a940bf (vllm-project#417). Keep the deferred Qwen3.5 NPU recipe absent and retain the validated 128-token GLM test limit. Both conflict resolutions preserve the current tree; no capability or validation scope is expanded. Signed-off-by: Meihan-chen <zr010426ztt@outlook.com>
92 tasks
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
enforce-eager: Removed theenforce-eagerparameter forQwen3.5-35B-A3Bonvime-ascendto enable optimized graph execution mode, reducing PyEager overhead and improving overall hardware execution efficiency on Ascend NPU.Qwen3.5-35B-A3Bbetweenvime-ascendandslime-ascend.check_mtp_only_gradfalse-positive failure in the GLM-4.7-Flash CI test, where the MTP isolation assertion incorrectly fired when GRPO group rewards differ (non-all-truncated case). Closes [NPU][Bug] check_mtp_only_grad fails on GLM-4.7-Flash #421.Test Environment
Performance Comparison (
vime-ascendvsslime-ascend)Conclusion
vime-ascendachieves 14.6% end-to-end step time reduction overslime-ascend. The most significant gains come from:Actor Train tok/s is lower in Vime (-12.8%), but the overall pipeline efficiency is better due to significantly faster rollout and weight synchronization phases.