[refactor] Refactoring AscendFusedMoE #1229
Merged
Conversation
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
Collaborator
|
Can you address the review comment from #1169 |
zzzzwwjj
force-pushed
the
main
branch
2 times, most recently
from
92c8998 to
494088d
Compare
Contributor
There was a problem hiding this comment.
why graph batch size starts from 4? this should be added into torchair config
Collaborator
Author
There was a problem hiding this comment.
This function is used to generate graph batch size automatically, if you want to control graph batch size, you can use torchair_graph_config.graph_batch_sizes to input it.
vllm_ascend/ops/fused_moe.py
Outdated
Contributor
There was a problem hiding this comment.
this slice should be controlled by if not self.torchair_graph_enabled or is_prefill:
Collaborator
Author
There was a problem hiding this comment.
It is unnecessary to do this.
Signed-off-by: zzzzwwjj <1183291235@qq.com>
Collaborator
|
@ganyi1996ppo @jianzs @ApsarasX Please also take a look |
realliujiaxu
approved these changes
Jun 17, 2025
wangxiyuan
approved these changes
Jun 17, 2025
Collaborator
There was a problem hiding this comment.
Dose this means, we run all the scenario with full tokens?
Collaborator
Author
There was a problem hiding this comment.
indeed, if you don't set either of graph_batch_sizes_init and graph_batch_sizes.
Collaborator
There was a problem hiding this comment.
Since we already support unbalanced mc2, why not just keep the bs1 graph here to miner the pressure of communication?
vllm_ascend/models/deepseek_dbo.py
Outdated
Collaborator
There was a problem hiding this comment.
Why there are only delete in dbo, have you verified the functionality of this?
vllm_ascend/models/deepseek_v2.py
Outdated
Collaborator
There was a problem hiding this comment.
Can you add comments here to illustrate the expected input and output shape of self.expert?
Collaborator
Author
There was a problem hiding this comment.
if you pass in shared_experts, self.expert will return a tuple: (router_hidden_states, shared_hidden_states), else will return a tensor of router_hidden_states.
Collaborator
Author
There was a problem hiding this comment.
I will add this note in code at next PR.
ganyi1996ppo
approved these changes
Jun 17, 2025
ApsarasX
approved these changes
Jun 17, 2025
Collaborator
|
The dbo path seems have some issue in this PR, just discussed with @zzzzwwjj , we quick merge this code first for further release procedure, the fix PR will be filed asap after this PR merged. |
Collaborator
|
@zzzzwwjj Please cherry-pick this PR to the dev branch |
ganyi1996ppo
pushed a commit
to ganyi1996ppo/vllm-ascend
that referenced
this pull request
Jun 17, 2025
<!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> This PR is used for resolved [issue 1147](vllm-project#1147) 1. Move fused_moe code into one file `fused_moe.py`. 2. Integrate branch conditions into function `get_fused_moe_state`. <!-- - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Fixes # --> 1. This PR has removed the env `VLLM_ENABLE_MC2`, because I think this env is useless, we can make judgments based on the current scenario without this env, it will only increase complexity. 2. This PR has removed the env `USING_LCCL_COM`, because this env has already expired. 3. `additional_config.expert_tensor_parallel_size` has already expired, and now we also use parameter `enable_expert_parallel`, consistent with the vLLM. <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> Signed-off-by: zzzzwwjj <1183291235@qq.com>
ganyi1996ppo
pushed a commit
to ganyi1996ppo/vllm-ascend
that referenced
this pull request
Jun 17, 2025
<!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> This PR is used for resolved [issue 1147](vllm-project#1147) 1. Move fused_moe code into one file `fused_moe.py`. 2. Integrate branch conditions into function `get_fused_moe_state`. <!-- - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Fixes # --> 1. This PR has removed the env `VLLM_ENABLE_MC2`, because I think this env is useless, we can make judgments based on the current scenario without this env, it will only increase complexity. 2. This PR has removed the env `USING_LCCL_COM`, because this env has already expired. 3. `additional_config.expert_tensor_parallel_size` has already expired, and now we also use parameter `enable_expert_parallel`, consistent with the vLLM. <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> Signed-off-by: zzzzwwjj <1183291235@qq.com> Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
Yikun
pushed a commit
that referenced
this pull request
Jun 17, 2025
### What this PR does / why we need it? This PR is the cherry-pick of the PR #1229 which have already merged into the main branch. This PR is used for resolved [issue 1147](#1147) 1. Move fused_moe code into one file `fused_moe.py`. 2. Integrate branch conditions into function `get_fused_moe_state`. ### Does this PR introduce _any_ user-facing change? 1. This PR has removed the env `VLLM_ENABLE_MC2`, because I think this env is useless, we can make judgments based on the current scenario without this env, it will only increase complexity. 2. This PR has removed the env `USING_LCCL_COM`, because this env has already expired. 3. `additional_config.expert_tensor_parallel_size` has already expired, and now we also use parameter `enable_expert_parallel`, consistent with the vLLM. ### How was this patch tested? CI passed Signed-off-by: zzzzwwjj <1183291235@qq.com> Signed-off-by: ganyi <pleaplusone.gy@gmail.com> Co-authored-by: zzzzwwjj <34335947+zzzzwwjj@users.noreply.github.com>
This was referenced Jun 18, 2025
This was referenced Jun 18, 2025
wangxiyuan
pushed a commit
that referenced
this pull request
Jun 25, 2025
### What this PR does / why we need it? Add `max_num_tokens_across_dp` to AscendMetadata to fix dp This pr fixes the bug introduced by #1229, which add an arg `max_num_tokens_across_dp` when dp_size > 1. Signed-off-by: MengqingCao <cmq0113@163.com>
weijinqian0
pushed a commit
to weijinqian0/vllm-ascend
that referenced
this pull request
Jun 30, 2025
### What this PR does / why we need it? Add `max_num_tokens_across_dp` to AscendMetadata to fix dp This pr fixes the bug introduced by vllm-project#1229, which add an arg `max_num_tokens_across_dp` when dp_size > 1. Signed-off-by: MengqingCao <cmq0113@163.com>
chopper0126
pushed a commit
to chopper0126/vllm-ascend
that referenced
this pull request
Oct 16, 2025
<!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> ### What this PR does / why we need it? This PR is used for resolved [issue 1147](vllm-project#1147) 1. Move fused_moe code into one file `fused_moe.py`. 2. Integrate branch conditions into function `get_fused_moe_state`. <!-- - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Fixes # --> ### Does this PR introduce _any_ user-facing change? 1. This PR has removed the env `VLLM_ENABLE_MC2`, because I think this env is useless, we can make judgments based on the current scenario without this env, it will only increase complexity. 2. This PR has removed the env `USING_LCCL_COM`, because this env has already expired. 3. `additional_config.expert_tensor_parallel_size` has already expired, and now we also use parameter `enable_expert_parallel`, consistent with the vLLM. <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> ### How was this patch tested? <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> Signed-off-by: zzzzwwjj <1183291235@qq.com>
chopper0126
pushed a commit
to chopper0126/vllm-ascend
that referenced
this pull request
Oct 16, 2025
### What this PR does / why we need it? Add `max_num_tokens_across_dp` to AscendMetadata to fix dp This pr fixes the bug introduced by vllm-project#1229, which add an arg `max_num_tokens_across_dp` when dp_size > 1. Signed-off-by: MengqingCao <cmq0113@163.com>
Angazenn
pushed a commit
to Angazenn/vllm-ascend
that referenced
this pull request
Oct 21, 2025
<!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> ### What this PR does / why we need it? This PR is used for resolved [issue 1147](vllm-project#1147) 1. Move fused_moe code into one file `fused_moe.py`. 2. Integrate branch conditions into function `get_fused_moe_state`. <!-- - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Fixes # --> ### Does this PR introduce _any_ user-facing change? 1. This PR has removed the env `VLLM_ENABLE_MC2`, because I think this env is useless, we can make judgments based on the current scenario without this env, it will only increase complexity. 2. This PR has removed the env `USING_LCCL_COM`, because this env has already expired. 3. `additional_config.expert_tensor_parallel_size` has already expired, and now we also use parameter `enable_expert_parallel`, consistent with the vLLM. <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> ### How was this patch tested? <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> Signed-off-by: zzzzwwjj <1183291235@qq.com>
Angazenn
pushed a commit
to Angazenn/vllm-ascend
that referenced
this pull request
Oct 21, 2025
### What this PR does / why we need it? Add `max_num_tokens_across_dp` to AscendMetadata to fix dp This pr fixes the bug introduced by vllm-project#1229, which add an arg `max_num_tokens_across_dp` when dp_size > 1. Signed-off-by: MengqingCao <cmq0113@163.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 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?
This PR is used for resolved issue 1147
fused_moe.py.get_fused_moe_state.Does this PR introduce any user-facing change?
VLLM_ENABLE_MC2, because I think this env is useless, we can make judgments based on the current scenario without this env, it will only increase complexity.USING_LCCL_COM, because this env has already expired.additional_config.expert_tensor_parallel_sizehas already expired, and now we also use parameterenable_expert_parallel, consistent with the vLLM.How was this patch tested?