[Kernel] add custom moe ops for prefill#4194
Merged
wangxiyuan merged 4 commits intovllm-project:mainfrom Dec 8, 2025
Merged
Conversation
Contributor
|
👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:
If CI fails, you can run linting and testing checks locally according Contributing and Testing. |
![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 introduces custom Mixture of Experts (MoE) operations for the prefill stage, which appears to be targeted for Ascend hardware. The changes include new kernel implementations, host-side logic, build scripts, and PyTorch bindings. While the core logic for the custom ops is complex and hardware-specific, I've identified several critical issues related to correctness in the PyTorch binding code, as well as high-severity issues in the build scripts and C++ host code that affect robustness and maintainability. Please address the critical correctness bugs and consider the high-severity suggestions to improve the quality of the code.
shiro-zzzz
force-pushed
the
add_moe_normal
branch
from
b4d3438 to
f5fcd76
Compare
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
shiro-zzzz
force-pushed
the
add_moe_normal
branch
from
f5fcd76 to
a9f39fd
Compare
shiro-zzzz
force-pushed
the
add_moe_normal
branch
from
a9f39fd to
1e77854
Compare
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
shiro-zzzz
force-pushed
the
add_moe_normal
branch
3 times, most recently
from
9f41ea7 to
366c7b4
Compare
shiro-zzzz
force-pushed
the
add_moe_normal
branch
6 times, most recently
from
e8180d5 to
c1b05d3
Compare
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
shiro-zzzz
force-pushed
the
add_moe_normal
branch
from
c1b05d3 to
8a9828f
Compare
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
shiro-zzzz
force-pushed
the
add_moe_normal
branch
from
8a9828f to
8972e55
Compare
1.Add the implementation of normal Aclnn operators: MoeCombineNormal, MoeDispatchNormal, NotifyDispatch，and DispatchLayout. 2.Provide PyTorch interfaces for the normal operators: get_dispatch_layout, dispatch_prefill, and combine_prefill. Signed-off-by: shiro-zzzz <zhangdianhao@huawei.com>
shiro-zzzz
force-pushed
the
add_moe_normal
branch
from
8972e55 to
4ece98d
Compare
wangxiyuan
approved these changes
Dec 8, 2025
MengqingCao
added a commit
that referenced
this pull request
Dec 8, 2025
This reverts commit 0617d7d.
wangxiyuan
added a commit
that referenced
this pull request
Dec 8, 2025
Reverts #4194 as it broke CI in https://github.com/vllm-project/vllm-ascend/actions/runs/20030369087/job/57437687382?pr=4791 Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
Collaborator
|
@shiro-zzzz this pr is reverted by #4806, plz fix the issue mentioned in #4806 and redo this, thx! |
weijinqian0
pushed a commit
to weijinqian0/vllm-ascend
that referenced
this pull request
Dec 9, 2025
### What this PR does / why we need it? 1.Add the implementation of normal Aclnn operators: MoeCombineNormal, MoeDispatchNormal, NotifyDispatch，and DispatchLayout. - MoeCombineNormal: Implements the combine logic within MoE operations. - MoeDispatchNormal: Implements the dispatch logic within MoE operations. - NotifyDispatch: Exchanges topk_idx information among different ranks to calculate the device memory required for the dispatch stage. - DispatchLayout: Used to calculate information related to the device memory layout for the dispatch stage. 2.Provide PyTorch interfaces for normal operators—get_dispatch_layout, dispatch_prefill, and combine_prefill—to be used for MoE communication during the prefill stage in vLLM. - get_dispatch_layout: Calculates information related to the device memory layout for the dispatch operator, and is called before dispatch_prefill. - dispatch_prefill: Initiates the dispatch operation. - combine_prefill: Initiates the combine operation. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? The functionality has already been validated using the local Qwen model. Test cases will be added after support for multi-NPU use cases in the CI pipeline is finalized. - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: shiro-zzzz <zhangdianhao@huawei.com> Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
weijinqian0
pushed a commit
to weijinqian0/vllm-ascend
that referenced
this pull request
Dec 9, 2025
Reverts vllm-project#4194 as it broke CI in https://github.com/vllm-project/vllm-ascend/actions/runs/20030369087/job/57437687382?pr=4791 Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com> Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Clorist33
pushed a commit
to Clorist33/vllm-ascend
that referenced
this pull request
Dec 10, 2025
### What this PR does / why we need it? 1.Add the implementation of normal Aclnn operators: MoeCombineNormal, MoeDispatchNormal, NotifyDispatch，and DispatchLayout. - MoeCombineNormal: Implements the combine logic within MoE operations. - MoeDispatchNormal: Implements the dispatch logic within MoE operations. - NotifyDispatch: Exchanges topk_idx information among different ranks to calculate the device memory required for the dispatch stage. - DispatchLayout: Used to calculate information related to the device memory layout for the dispatch stage. 2.Provide PyTorch interfaces for normal operators—get_dispatch_layout, dispatch_prefill, and combine_prefill—to be used for MoE communication during the prefill stage in vLLM. - get_dispatch_layout: Calculates information related to the device memory layout for the dispatch operator, and is called before dispatch_prefill. - dispatch_prefill: Initiates the dispatch operation. - combine_prefill: Initiates the combine operation. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? The functionality has already been validated using the local Qwen model. Test cases will be added after support for multi-NPU use cases in the CI pipeline is finalized. - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: shiro-zzzz <zhangdianhao@huawei.com>
Clorist33
pushed a commit
to Clorist33/vllm-ascend
that referenced
this pull request
Dec 10, 2025
Reverts vllm-project#4194 as it broke CI in https://github.com/vllm-project/vllm-ascend/actions/runs/20030369087/job/57437687382?pr=4791 Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
Mercykid-bash
pushed a commit
to Mercykid-bash/vllm-ascend
that referenced
this pull request
Dec 10, 2025
### What this PR does / why we need it? 1.Add the implementation of normal Aclnn operators: MoeCombineNormal, MoeDispatchNormal, NotifyDispatch，and DispatchLayout. - MoeCombineNormal: Implements the combine logic within MoE operations. - MoeDispatchNormal: Implements the dispatch logic within MoE operations. - NotifyDispatch: Exchanges topk_idx information among different ranks to calculate the device memory required for the dispatch stage. - DispatchLayout: Used to calculate information related to the device memory layout for the dispatch stage. 2.Provide PyTorch interfaces for normal operators—get_dispatch_layout, dispatch_prefill, and combine_prefill—to be used for MoE communication during the prefill stage in vLLM. - get_dispatch_layout: Calculates information related to the device memory layout for the dispatch operator, and is called before dispatch_prefill. - dispatch_prefill: Initiates the dispatch operation. - combine_prefill: Initiates the combine operation. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? The functionality has already been validated using the local Qwen model. Test cases will be added after support for multi-NPU use cases in the CI pipeline is finalized. - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: shiro-zzzz <zhangdianhao@huawei.com>
Mercykid-bash
pushed a commit
to Mercykid-bash/vllm-ascend
that referenced
this pull request
Dec 10, 2025
Reverts vllm-project#4194 as it broke CI in https://github.com/vllm-project/vllm-ascend/actions/runs/20030369087/job/57437687382?pr=4791 Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
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
3 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.
What this PR does / why we need it?
1.Add the implementation of normal Aclnn operators: MoeCombineNormal, MoeDispatchNormal, NotifyDispatch，and DispatchLayout.
2.Provide PyTorch interfaces for normal operators—get_dispatch_layout, dispatch_prefill, and combine_prefill—to be used for MoE communication during the prefill stage in vLLM.
Does this PR introduce any user-facing change?
No
How was this patch tested?
The functionality has already been validated using the local Qwen model. Test cases will be added after support for multi-NPU use cases in the CI pipeline is finalized.