Skip to content

[Misc]feat: adapt to vLLM main (0d4d334e...78e7a7b9) - #251

Draft
Meihan-chen wants to merge 41 commits into
nv-action:mainfrom
Meihan-chen:main2main_auto_2026-05-20_01-34
Draft

[Misc]feat: adapt to vLLM main (0d4d334e...78e7a7b9)#251
Meihan-chen wants to merge 41 commits into
nv-action:mainfrom
Meihan-chen:main2main_auto_2026-05-20_01-34

Conversation

@Meihan-chen

@Meihan-chen Meihan-chen commented May 20, 2026 •
edited by github-actions Bot
Loading

Copy link
Copy Markdown
Collaborator

Main2Main Summary

Status: completed
Upstream range: 0d4d334eaa583b9c09aa4eb7538c22db99fd84b3..78e7a7b9b0b9c285bf6978c3fc09eeecea3ff230
Reached upstream commit: 78e7a7b9b0b9c285bf6978c3fc09eeecea3ff230
Steps: 10/10
CI suite: e2e-singlecard-light, e2e-2card-light, e2e-4card-light

Result

Successfully upgraded vllm-ascend from upstream vLLM commit 0d4d334eaa583b9c09aa4eb7538c22db99fd84b3 to target commit 78e7a7b9b0b9c285bf6978c3fc09eeecea3ff230 (80 commits across 10 steps). All steps passed CI verification without requiring code adaptation beyond commit reference updates. The upstream changes were predominantly CUDA/ROCm/CPU/XPU-specific optimizations and features that do not affect vllm-ascend's Ascend NPU implementations.

Completed Steps

Step Upstream range vllm-ascend commit CI result Summary
step-1 0d4d334ea..4b364f810e ac9ee8f passed InternS2 model support, SWA prefix-cache optimization - commit reference only
step-2 4b364f810e..75fd68c7a be3f353 passed PoolingOfflineMixin split - commit reference only
step-3 75fd68c7a..491e8d8539 3e8e5aa passed IPC weight sync, MLA bmm optimization - commit reference only
step-4 491e8d8539..b2c58ee942 d99a003 passed IR op priority at worker init, FP32 gumbel - commit reference only
step-5 b2c58ee942..32b717790 6dd7370 passed LinearBackend arg, LoRA dedup - commit reference only
step-6 32b717790..787bc0d031 bafc635 passed Breakable CUDA graph experimental - commit reference only
step-7 787bc0d031..03ddc1c9bc fb8e4b0 passed Mooncake disk offloading, TritonFP8MoE - commit reference only
step-8 03ddc1c9bc..998714b21b 40bcf3c passed LoRA 2D/3D MoE support, encoder cudagraph - commit reference only
step-9 998714b21b..e414e1f1c0 382bb1b passed CPU MXFP4 MoE, KV offload bugfix - commit reference only
step-10 e414e1f1c0..78e7a7b9b0 935b160 passed AWQ Marlin MoE refactor, mamba speculative fix - commit reference only

Changes Made

  • Updated vLLM commit reference from 0d4d334eaa583b9c09aa4eb7538c22db99fd84b3 to 78e7a7b9b0b9c285bf6978c3fc09eeecea3ff230 in:
    • docs/source/conf.py (main_vllm_commit)
    • .github/workflows/dockerfiles/Dockerfile.lint
    • .github/workflows/pr_test_full.yaml
    • .github/workflows/pr_test_light.yaml
  • No vllm-ascend code adaptations required - all upstream changes were platform-specific (CUDA, ROCm, CPU, XPU) or backward-compatible interface additions
  • No version compatibility guards needed to be added

CI Verification

  • Passed: All 10 steps passed e2e-singlecard-light, e2e-2card-light, e2e-4card-light suites
  • Treated as env flakes: none
  • Last successful step: step-10 (final)

Key Observations

  • All 80 upstream commits absorbed across 10 steps without breaking vllm-ascend
  • Upstream changes were well-isolated to CUDA/ROCm/CPU-specific implementations
  • vllm-ascend's Ascend-specific implementations (MLA, MoE, attention, workers) remain unaffected
  • Version compatibility guards using vllm_version_is("0.20.2") remain valid (main_vllm_tag unchanged)
  • 10 signed commits created in vllm-ascend repository

Commits Created (vllm-ascend)

  1. ac9ee8f532ca269df7c4c5984a7d428fb54e5890
  2. be3f35319a08db7f5182ab09f30a4120e3701ebd
  3. 3e8e5aa75837ba7b5bcc7168c195f2c553e1d7e4
  4. d99a0035096db52d05c21bdfd00de193586c0d51
  5. 6dd7370c12f0ef48ba2437d0f4583cbab2c4b589
  6. bafc63555d617651da14aebfbf76fb4e96572d91
  7. fb8e4b0975934fe965017ba2a34e1a7d6759c210
  8. 40bcf3c437917a81c1f7d32d202883faf13b05c8
  9. 382bb1b894ea54db5b19e13d17632596aeb0e57c
  10. 935b160603b125f3940b61d7f2a88c2b4ecf0edf

Meihan-chen and others added 30 commits May 20, 2026 09:26
@Meihan-chen
fix yaml
6fb5bf7 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
main2main_auto
043c3bf 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
main2main_auto_test
1021701 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
refactor bisect
b67abd5 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
refactor issue review
cd68f3f 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
docs: add workflow-native main2main orchestration spec
f5b9cb2
@Meihan-chen
refactor: migrate main2main to workflow-native orchestration
779988b
@Meihan-chen
fix: harden workflow-native main2main guards
7d9ea2e
@Meihan-chen
fix
7d8e5f6 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
fix 2
1b61941 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
fix 3
858a118 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
fix 3
0f11084 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
feat: add simplified main2main workflows
f958b32
@zhangxinyuehfad @Meihan-chen
fix doc
96f9c94 
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
@Meihan-chen
main2main_auto_fix
465e638 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@zhangxinyuehfad @Meihan-chen
fix doc
21848e1 
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
@Meihan-chen
main2main_auto_fix
ffb975d 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
main2main_auto_fix
eb90517 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
main2main_mock
57efb21 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
main2main_refactor
cb2c5be 
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
@Meihan-chen
fix skill
a2ad443 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
fix main2main
9676c6f 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main auto
ba74271 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main skill
1ec2226 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main auto
b68dd28 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main auto claude action
ff9da63 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main auto claude cli
952d021 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main auto
6c2fc68 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main skill
56ecd45 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@Meihan-chen
main2main skill
b6bad79 
Signed-off-by: Meihan-chen <239345480+Meihan-chen@users.noreply.github.com>
@zhangxinyuehfad
zhangxinyuehfad force-pushed the main branch 4 times, most recently from 05b98c1 to 38085d5 Compare June 26, 2026 09:09
@zhangxinyuehfad
zhangxinyuehfad force-pushed the main branch 6 times, most recently from 1cd9ff4 to b7570c6 Compare July 7, 2026 03:28
@zhangxinyuehfad
zhangxinyuehfad force-pushed the main branch 4 times, most recently from 0c96baa to e7d055b Compare July 13, 2026 09:44
xqchen7 pushed a commit that referenced this pull request Jul 15, 2026
@Yikun @xqchen7
[CI] Add dispatch job to leverage dynamic devices (#251)
a137b2d 
### What this PR does / why we need it?
Add dispatch job to leverage jobs to dynamic devices include 2 stage as
below:

The dispatch job will spend extra about `10s * parallel number + 30s`
time to wait other job launch container and release lock.

- **Stage 1: Acquire lock**
add a dispatch job, this job use lockfile to acquire locks and then get
device number dynamically
- **Stage 2.1: Launch container with dynamic device**
pass the device number via output and start the container job with
dynamic device
- **Stage 2.2: Release lock**
once the job started, release the lock.

In the backend, we use multiple path to setup multiple self host runners
as load balancer:
```
$ pwd
/home/action
$ ll | grep actions
drwx------   6 action action 4096 Mar  7 08:55 actions-runner-01
drwx------   6 action action 4096 Mar  7 08:55 actions-runner-02
drwx------   6 action action 4096 Mar  7 08:55 actions-runner-03
drwx------   6 action action 4096 Mar  7 08:56 actions-runner-04
drwx------   4 action action 4096 Jan 24 22:08 actions-runner-05
drwx------   4 action action 4096 Jan 24 22:08 actions-runner-06
```

```
adduser -G docker action
su action
pip3 install docker prettytable
sudo yum install procmail
```

### Does this PR introduce _any_ user-facing change?
NO

### How was this patch tested?
- CI passed
- E2E test manully, triggered 3 jobs in parallel:
- [1st
job](https://github.com/vllm-project/vllm-ascend/actions/runs/13711345757/job/38348309297)
dispatch to /dev/davinci2.
- [2nd
job](https://github.com/vllm-project/vllm-ascend/actions/runs/13711348739/job/38348316250)
dispatch to /dev/davinci3
- [3rd
job](https://github.com/vllm-project/vllm-ascend/actions/runs/13711351493/job/38348324551)
dispatch to /dev/davinci4

Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
Signed-off-by: xqchen7 <chenxueqing7@huawei.com>
@MrZ20
MrZ20 force-pushed the main branch 3 times, most recently from 8899462 to ce1086a Compare July 24, 2026 12:43
@zhangxinyuehfad
zhangxinyuehfad force-pushed the main branch 10 times, most recently from 52fad8d to e594504 Compare August 6, 2026 07:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@wangxiyuan wangxiyuan Awaiting requested review from wangxiyuan wangxiyuan will be requested when the pull request is marked ready for review wangxiyuan is a code owner
@Yikun Yikun Awaiting requested review from Yikun Yikun will be requested when the pull request is marked ready for review Yikun is a code owner

Assignees

No one assigned

Labels

ci/build documentation Improvements or additions to documentation merge-conflicts module:tests module:tools

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@Meihan-chen @zhangxinyuehfad