[Feat] chunkprefill mla support torchair graph #1772

haojiangzheng · 2025-07-14T04:50:10Z

What this PR does / why we need it?

chunkprefill mla only support eager mode now，we want to optimaze it by support torchair graph, the idea is simpler, when all the request is running in decode, use torchair graph to deal with it, else when chunkprefill or prefill only, use the eager mode

Does this PR introduce any user-facing change?

we can start the serve with chunkprefill mla torchair graph like this:

export VLLM_USE_V1=1
export VLLM_VERSION=0.9.0
python -m vllm.entrypoints.openai.api_server --model=DeepSeek-R1-W8A8-VLLM
--quantization ascend
--trust-remote-code
--distributed-executor-backend=mp
--port 8006
-tp=8
-dp=2
--max-num-seqs 24
--max-model-len 32768
--max-num-batched-tokens 8192
--block-size 128
--additional-config '{"torchair_graph_config":{"enabled":true,"use_cached_graph":true,"graph_batch_sizes":[24]},"expert_tensor_parallel_size":16}'
--gpu-memory-utilization 0.96

How was this patch tested?

vLLM version: v0.10.0
vLLM main: vllm-project/vllm@ebf7605

ApsarasX · 2025-07-14T06:22:09Z

Missing the fix code for ChunkPrefill OOM

haojiangzheng · 2025-07-18T03:38:52Z

Missing the fix code for ChunkPrefill OOM

fixed

codecov · 2025-07-22T07:10:35Z

Codecov Report

❌ Patch coverage is 14.28571% with 12 lines in your changes missing coverage. Please review.
✅ Project coverage is 76.22%. Comparing base (1ab1541) to head (3863246).
⚠️ Report is 5 commits behind head on main.

Files with missing lines	Patch %	Lines
vllm_ascend/attention/mla_v1.py	7.69%	12 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1772      +/-   ##
==========================================
- Coverage   76.31%   76.22%   -0.09%     
==========================================
  Files         116      117       +1     
  Lines       13238    13273      +35     
==========================================
+ Hits        10102    10117      +15     
- Misses       3136     3156      +20

Flag	Coverage Δ
unittests	`76.22% <14.28%> (-0.09%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

github-actions · 2025-07-26T09:23:50Z

This pull request has conflicts, please resolve those before we can evaluate the pull request.

github-actions · 2025-07-31T12:10:59Z

This pull request has conflicts, please resolve those before we can evaluate the pull request.

github-actions · 2025-08-06T11:39:08Z

This pull request has conflicts, please resolve those before we can evaluate the pull request.

Signed-off-by: haojiangzheng <justineric096@gmail.com>

* enable mm allreduce test (vllm-project#2192) ### What this PR does / why we need it? This PR is to add e2e test for using npu_mm_all_reduce_base fusion kernel. ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? not involved - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@5d5d419 Signed-off-by: Ronald1995 <ronaldautomobile@163.com> * [main] remove torch.cat and replace it by List[0] (vllm-project#2153) ### What this PR does / why we need it? torch_npu.npu_grouped_matmul: https://www.hiascend.com/document/detail/zh/Pytorch/710/apiref/torchnpuCustomsapi/context/torch_npu-npu_grouped_matmul.md According to the document, when `split_item` is 2 or 3, `torch_npu.npu_grouped_matmul` will return a list which has one element. Therefore, the `torch.cat` after `torch_npu.npu_grouped_matmul` is unnecessary. ### Does this PR introduce _any_ user-facing change? not involved ### How was this patch tested? ut and e2e covered: `tests/ut/ops/test_fused_ops.py`, `tests/e2e/singlecard/ops/test_fused_moe.py` **performance**: (qwen3 30B, 2k->20k) base: Total Token throughput (tok/s): 667.76 remove cat: Total Token throughput (tok/s): 680.82 - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@fa00c5d Signed-off-by: huangxialu <huangxialu1@huawei.com> * [CI][Quickfix] Fix AscendFusedMoE init error (vllm-project#2268) ### What this PR does / why we need it? Fix AscendFusedMoE init error. Use `super().__init__()` instead of `super(FusedMoE, self).__init__()` to ensure the member variables in base class could be called by the children class ### Does this PR introduce _any_ user-facing change? N/A ### How was this patch tested? CI passed with new existing test. - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@766bc81 --------- Signed-off-by: MengqingCao <cmq0113@163.com> * Fix accuracy test config and add DeepSeek-V2-Lite test (vllm-project#2261) ### What this PR does / why we need it? This PR fix accuracy test related to vllm-project#2073, users can now perform accuracy tests on multiple models simultaneously and generate different report files by running: ```bash cd ~/vllm-ascend pytest -sv ./tests/e2e/models/test_lm_eval_correctness.py \ --config-list-file ./tests/e2e/models/configs/accuracy.txt ``` ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? <img width="1648" height="511" alt="image" src="https://github.com/user-attachments/assets/1757e3b8-a6b7-44e5-b701-80940dc756cd" /> - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@766bc81 --------- Signed-off-by: Icey <1790571317@qq.com> * Fix accuracy test create PR (vllm-project#2274) ### What this PR does / why we need it? Fix create PR of accuracy test ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? Local testing: nv-action/vllm-benchmarks#87 - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@099c046 --------- Signed-off-by: Icey <1790571317@qq.com> * Add ut for test_communicator.py (vllm-project#2293) ### What this PR does / why we need it? Add ut for test_communicator.py - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@e5ebeeb Signed-off-by: yangqinghao-cmss <yangqinghao_yewu@cmss.chinamobile.com> * [CI] Fix broken CI (vllm-project#2302) 1. disable test_eagle_ccorrectness test, we'll reopen it once oom error fixed. 2. drop transformers version limit for main, since vLLM rely on >=4.55.0, see: vllm-project/vllm@65552b4 3. fix kv_connector_output bug, see: vllm-project/vllm@796bae0 - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@d1af8b7 Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> * [2/N][Refactor] torchair model runner refactor (vllm-project#2204) There is lot of torchair code in model runner leading the code hard for maintenance. We'll create new torchair_model_runner to split torchair related logic. Following the workflow vllm-project#2203 What's this PR do: move `torchair` related logic into `_get_forward_metadata_across_dp` and override it in torchair model runner - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@1b99028 Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> * [core] Support capture custom ops into aclgraph (vllm-project#2113) ### What this PR does / why we need it? Thanks to the PR vllm-project#426 make vllm-ascend support the aclgraph inference to reduce the host overhead. However, the capability of aclgraph strongly relies on the functionality provided by `torch.compile`, which is the key feature supported in torch 2.x . Therefore, capture custom op into aclgraph is only possible when it can be recognize and captured by `torch.compile`. In this PR, we register the meta implementation of current custom ops to enable the fx graph capture. And by doing that, insert those custom ops into aclgraph become a natural thing to the ascend runtime. ### Does this PR introduce _any_ user-facing change? No user face change. ### How was this patch tested? Tested in unittest, we will integrate the `rotary_embedding` op into a small custom model and use `torch.compile` and aclgraph to capture and replay it to verify its functionality. - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@1b99028 --------- Signed-off-by: ganyi <pleaplusone.gy@gmail.com> * Bump actions/download-artifact from 4 to 5 (vllm-project#2311) Bumps [actions/download-artifact](https://github.com/actions/download-artifact) from 4 to 5. - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@ebf7605 Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * [Perf][MTP] Optimize reject sampler in greedy situation. (vllm-project#2137) This PR port optimization in PR vllm-project#2002 to main and makes it cleaner. - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@afa5b7c --------- Signed-off-by: whx-sjtu <2952154980@qq.com> * [3/N][Refactor] torchair model runner refactor (vllm-project#2207) There is lot of torchair code in model runner leading the code hard for maintenance. We'll create new torchair_model_runner to split torchair related logic. Following the workflow vllm-project#2203, this is the first PR. What's this PR do: create common function `_build_attention_metadata` and `_generate_dummy_run_hidden_states` for dummy_run - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@ebf7605 Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> * [Feat] chunkprefill mla support torchair graph (vllm-project#1772) chunkprefill mla only support eager mode now，we want to optimaze it by support torchair graph, the idea is simple, when all the request is running in decode, use torchair graph to deal with it, else when chunkprefill or prefill only, use the eager mode - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@ebf7605 Signed-off-by: haojiangzheng <justineric096@gmail.com> Co-authored-by: haojiangzheng <justineric096@gmail.com> * [4/N][Refactor] torchair model runner refactor (vllm-project#2208) There is lot of torchair code in model runner leading the code hard for maintenance. We'll create new torchair_model_runner to split torchair related logic. Following the workflow vllm-project#2203, this is the first PR. What's this PR do: create common function `_convert_torch_foramt` for initialize_kv_cache - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@14a5d90 Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> * Configure Gemini (vllm-project#2298) ### What this PR does / why we need it? This PR requests Gemini AI to review PRs. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? NA - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@14a5d90 Signed-off-by: QwertyJack <7554089+QwertyJack@users.noreply.github.com> * ut: add ci guard for ut coverage (vllm-project#2317) ### What this PR does / why we need it? add ci guard for ut coverage, if ut coverage of patch pr is below 80%, the ci will failed/ ### Does this PR introduce _any_ user-facing change? not involved ### How was this patch tested? not involved - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@458e74e --------- Signed-off-by: Ronald1995 <ronaldautomobile@163.com> * [main][prefill optimization] Optimize parallel strategies to reduce communication overhead (vllm-project#2198) ### What this PR does / why we need it? 1.Shared Expert Sharding Strategy Update: Switched from TP-aligned to pure DP for shared experts, enabling more efficient execution. 2.O_Proj AllReduce → ReduceScatter: Reduced communication overhead by using ReduceScatter, made possible by pure DP sharding. 3.AllGather Postponed: Delayed to after QKV down projection to reduce synchronization impact during prefill. ### How was this patch tested? Adding ut case in `tests/ut/attention/test_mla_v1.py` #### How to run use parameter `--additional_config='{"enable_shared_expert_dp": true}'` ##### a.How to run eager mode eg: python -m vllm.entrypoints.openai.api_server --model=/model_path --trust-remote-code -tp 8 -dp 2 --enable_expert_parallel --port 8002 --max-model-len 5120 --max-num-batched-tokens 16384 --enforce-eager --disable-log-requests --additional_config='{"ascend_scheduler_config":{"enabled":true},"enable_shared_expert_dp": true,"chunked_prefill_for_mla":true}' ##### b.How to run graph mode eg: python -m vllm.entrypoints.openai.api_server --model=/model_path --trust-remote-code -tp 8 -dp 2 --enable_expert_parallel --port 8002 --max-model-len 5120 --max-num-batched-tokens 16384 --disable-log-requests --additional_config='{"ascend_scheduler_config":{"enabled":true},"enable_shared_expert_dp": true,"chunked_prefill_for_mla":true,"torchair_graph_config":{"enabled":true}}' - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@9edd1db --------- Signed-off-by: Wang Kunpeng <1289706727@qq.com> Signed-off-by: SlightwindSec <slightwindsec@gmail.com> Co-authored-by: SlightwindSec <slightwindsec@gmail.com> * [Doc] Update faq (vllm-project#2334) ### What this PR does / why we need it? - update determinitic calculation - update support device ### Does this PR introduce _any_ user-facing change? - Users should update ray and protobuf when using ray as distributed backend - Users should change to use `export HCCL_DETERMINISTIC=true` when enabling determinitic calculation ### How was this patch tested? N/A - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@ea1292a Signed-off-by: MengqingCao <cmq0113@163.com> * [5/N][Refactor] torchair model runner refactor (vllm-project#2216) There is lot of torchair code in model runner leading the code hard for maintenance. We'll create new torchair_model_runner to split torchair related logic. Following the workflow vllm-project#2203 What's this PR do: create common function `_capture_model` for capture_model - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@1891a26 Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> * [1/N][Feat] Support MoE models with ACL Graph and refactor MoE communication logic (vllm-project#2125) ### What this PR does / why we need it? This PR refactors the MoE (Mixture of Experts) communication logic by introducing a strategy pattern. It defines an abstract base class, `MoECommMethod`, which encapsulates different communication strategies for MoE layers. By decoupling the MoE implementation from any single communication method, this change makes it simpler to add, replace, or optimize communication strategies in the future. Plan / Roadmap 1. Introduce `MoECommMethod`, implement `AllGatherImpl`, and adapt ACL Graph handling to cover all scenarios (this PR). 2. Implement `MC2CommImpl` and `AllToAllCommImpl` to optimize performance in specific scenarios. 3. Enable W8A8 / Int8 models to use `unified_fused_experts`. Other notes * Data-parallel (DP) communication currently does not work with vLLM's dispatch/combine mechanisms; an alternative approach is required to resolve this incompatibility. - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@f7ad6a1 --------- Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com> * [Doc] Add container image save/load FAQ for offline environments (vllm-project#2347) ### What this PR does / why we need it? Add Docker export/import guide for air-gapped environments ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? NA - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@d16aa3d Signed-off-by: QwertyJack <7554089+QwertyJack@users.noreply.github.com> * [Bugfix] fix the oom when chunkprefill with long context like 64k (vllm-project#2319) The attn mask was declared in the mla.py，we don't need the splitfuse mask when mla chunkprefill, and this mask will cause memory problem when long context like 64k or 128k - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@14a5d90 --------- Signed-off-by: haojiangzheng <justineric096@gmail.com> --------- Signed-off-by: Ronald1995 <ronaldautomobile@163.com> Signed-off-by: huangxialu <huangxialu1@huawei.com> Signed-off-by: MengqingCao <cmq0113@163.com> Signed-off-by: Icey <1790571317@qq.com> Signed-off-by: yangqinghao-cmss <yangqinghao_yewu@cmss.chinamobile.com> Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> Signed-off-by: ganyi <pleaplusone.gy@gmail.com> Signed-off-by: dependabot[bot] <support@github.com> Signed-off-by: whx-sjtu <2952154980@qq.com> Signed-off-by: haojiangzheng <justineric096@gmail.com> Signed-off-by: QwertyJack <7554089+QwertyJack@users.noreply.github.com> Signed-off-by: Wang Kunpeng <1289706727@qq.com> Signed-off-by: SlightwindSec <slightwindsec@gmail.com> Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com> Co-authored-by: Ronald1995 <ronaldautomobile@163.com> Co-authored-by: huangxialu <huangxialu1@huawei.com> Co-authored-by: Mengqing Cao <cmq0113@163.com> Co-authored-by: Icey <1790571317@qq.com> Co-authored-by: yangqinghao-cmss <yangqinghao_yewu@cmss.chinamobile.com> Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com> Co-authored-by: Pleaplusone <pleaplusone.gy@gmail.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: whx <56632993+whx-sjtu@users.noreply.github.com> Co-authored-by: zhenghaojiang <zhjoneson@163.com> Co-authored-by: haojiangzheng <justineric096@gmail.com> Co-authored-by: jack <QwertyJack@users.noreply.github.com> Co-authored-by: Wang Kunpeng <1289706727@qq.com> Co-authored-by: SlightwindSec <slightwindsec@gmail.com> Co-authored-by: yiz-liu <136800916+yiz-liu@users.noreply.github.com>

chunkprefill mla only support eager mode now，we want to optimaze it by support torchair graph, the idea is simple, when all the request is running in decode, use torchair graph to deal with it, else when chunkprefill or prefill only, use the eager mode - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@ebf7605 Signed-off-by: haojiangzheng <justineric096@gmail.com> Co-authored-by: haojiangzheng <justineric096@gmail.com>

haojiangzheng force-pushed the main branch from 771611f to 5522e85 Compare July 18, 2025 03:34

haojiangzheng force-pushed the main branch 7 times, most recently from 68474e4 to 1c68685 Compare July 22, 2025 06:52

haojiangzheng force-pushed the main branch 2 times, most recently from 1c68685 to 0666d8f Compare July 22, 2025 08:39

ApsarasX added the ready read for review label Jul 24, 2025

github-actions bot added merge-conflicts and removed ready read for review labels Jul 26, 2025

haojiangzheng closed this Jul 30, 2025

haojiangzheng force-pushed the main branch from 0666d8f to d3c6dd9 Compare July 30, 2025 02:46

haojiangzheng reopened this Jul 30, 2025

github-actions bot removed the merge-conflicts label Jul 30, 2025

haojiangzheng force-pushed the main branch from 67be58e to b327992 Compare July 30, 2025 06:55

github-actions bot added the module:tests label Jul 30, 2025

haojiangzheng force-pushed the main branch from b327992 to 11df6ad Compare July 30, 2025 11:36

haojiangzheng changed the title ~~chunkprefill mla support torchair graph~~ [Feat] chunkprefill mla support torchair graph Jul 31, 2025

haojiangzheng force-pushed the main branch from 11df6ad to a021047 Compare July 31, 2025 10:44

github-actions bot added the merge-conflicts label Jul 31, 2025

haojiangzheng closed this Aug 1, 2025

haojiangzheng force-pushed the main branch from a021047 to d9f82eb Compare August 1, 2025 02:59

haojiangzheng reopened this Aug 1, 2025

github-actions bot removed the merge-conflicts label Aug 1, 2025

haojiangzheng force-pushed the main branch from 938631d to dcd3b88 Compare August 1, 2025 07:11

jianzs approved these changes Aug 4, 2025

View reviewed changes

haojiangzheng force-pushed the main branch 3 times, most recently from 8069fef to 853efb9 Compare August 5, 2025 06:39

github-actions bot added the merge-conflicts label Aug 6, 2025

haojiangzheng closed this Aug 7, 2025

haojiangzheng force-pushed the main branch from 055205c to 75e28d0 Compare August 7, 2025 03:17

haojiangzheng reopened this Aug 7, 2025

github-actions bot added merge-conflicts and removed module:tests merge-conflicts labels Aug 7, 2025

haojiangzheng force-pushed the main branch from 2da980e to 41730d3 Compare August 7, 2025 03:27

github-actions bot added module:tests and removed merge-conflicts labels Aug 7, 2025

chunkprefill mla with torchair graph

593fb1b

Signed-off-by: haojiangzheng <justineric096@gmail.com>

haojiangzheng force-pushed the main branch from 41730d3 to 593fb1b Compare August 7, 2025 04:26

Merge branch 'vllm-project:main' into main

3863246

jianzs merged commit eb43a47 into vllm-project:main Aug 11, 2025
25 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Feat] chunkprefill mla support torchair graph #1772

[Feat] chunkprefill mla support torchair graph #1772

Uh oh!

haojiangzheng commented Jul 14, 2025 •

edited by github-actions bot

Loading

Uh oh!

ApsarasX commented Jul 14, 2025

Uh oh!

haojiangzheng commented Jul 18, 2025

Uh oh!

codecov bot commented Jul 22, 2025 •

edited

Loading

Uh oh!

github-actions bot commented Jul 26, 2025

Uh oh!

github-actions bot commented Jul 31, 2025

Uh oh!

github-actions bot commented Aug 6, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

[Feat] chunkprefill mla support torchair graph #1772

[Feat] chunkprefill mla support torchair graph #1772

Uh oh!

Conversation

haojiangzheng commented Jul 14, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this PR does / why we need it?

Does this PR introduce any user-facing change?

How was this patch tested?

Uh oh!

ApsarasX commented Jul 14, 2025

Uh oh!

haojiangzheng commented Jul 18, 2025

Uh oh!

codecov bot commented Jul 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

github-actions bot commented Jul 26, 2025

Uh oh!

github-actions bot commented Jul 31, 2025

Uh oh!

github-actions bot commented Aug 6, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

haojiangzheng commented Jul 14, 2025 •

edited by github-actions bot

Loading

codecov bot commented Jul 22, 2025 •

edited

Loading