[Refactor] Collect scattered w8a8-dynamic quantization operations#1111
Closed
sdmyzlp wants to merge 2 commits intovllm-project:mainfrom
Closed
[Refactor] Collect scattered w8a8-dynamic quantization operations#1111sdmyzlp wants to merge 2 commits intovllm-project:mainfrom
sdmyzlp wants to merge 2 commits intovllm-project:mainfrom
Conversation
github-actions
bot
added
documentation
sdmyzlp
changed the title
sdmyzlp
force-pushed
the
br_quant_refactor
branch
from
5ed12bb to
9b37507
Compare
github-actions
bot
removed
documentation
sdmyzlp
force-pushed
the
br_quant_refactor
branch
3 times, most recently
from
4192815 to
542bd18
Compare
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
sdmyzlp
force-pushed
the
br_quant_refactor
branch
from
542bd18 to
e82a6ce
Compare
sdmyzlp
force-pushed
the
br_quant_refactor
branch
from
e82a6ce to
d65186f
Compare
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
sdmyzlp
force-pushed
the
br_quant_refactor
branch
from
d65186f to
edb6680
Compare
sdmyzlp
force-pushed
the
br_quant_refactor
branch
2 times, most recently
from
e5b5e83 to
976e1a7
Compare
Contributor
Author
|
@Yikun @wangxiyuan @ganyi1996ppo Please take a look at this. |
Collaborator
|
there is a w8a8 weight already. https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V2-Lite-W8A8 Can you take a try? |
sdmyzlp
force-pushed
the
br_quant_refactor
branch
from
976e1a7 to
c0acc46
Compare
sdmyzlp
force-pushed
the
br_quant_refactor
branch
7 times, most recently
from
41a3600 to
a4200e0
Compare
Contributor
Author
E2E testcase added. @wangxiyuan |
AscendW8A8DynamicLinearMethod is integrated into CustomDeepseekV2MLP in a very awkward way, causing scattered quantization operations all over the model scripts. Refactor to solve this problem. Signed-off-by: sdmyzlp <lrwei2@petalmail.com>
The model of chosen is vllm-ascend/DeepSeek-V2-Lite-W8A8. Signed-off-by: sdmyzlp <lrwei2@petalmail.com>
sdmyzlp
force-pushed
the
br_quant_refactor
branch
from
a4200e0 to
2a10573
Compare
ganyi1996ppo
pushed a commit
that referenced
this pull request
Jun 11, 2025
Contains on #1111 for completeness. <!-- 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? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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? No. <!-- 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? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com>
Contributor
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
Contributor
Author
|
Closed due to its super set #997 has been merged. |
raindaywhu
pushed a commit
to raindaywhu/vllm-ascend
that referenced
this pull request
Jun 16, 2025
Contains on vllm-project#1111 for completeness. <!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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 # --> No. <!-- 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. --> Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com>
momo609
pushed a commit
to momo609/vllm-ascend
that referenced
this pull request
Jun 17, 2025
Contains on vllm-project#1111 for completeness. <!-- 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? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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? No. <!-- 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? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com>
momo609
pushed a commit
to momo609/vllm-ascend
that referenced
this pull request
Jun 17, 2025
Contains on vllm-project#1111 for completeness. <!-- 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? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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? No. <!-- 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? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com> Signed-off-by: wangxiaoxin (A) <wangxiaoxin7@huawei.com>
momo609
pushed a commit
to momo609/vllm-ascend
that referenced
this pull request
Jun 17, 2025
Contains on vllm-project#1111 for completeness. <!-- 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? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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? No. <!-- 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? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com> Signed-off-by: wangxiaoxin (A) <wangxiaoxin7@huawei.com>
shiyuan680
pushed a commit
to raindaywhu/vllm-ascend
that referenced
this pull request
Jul 7, 2025
Contains on vllm-project#1111 for completeness. <!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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 # --> No. <!-- 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. --> Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com>
chopper0126
pushed a commit
to chopper0126/vllm-ascend
that referenced
this pull request
Oct 16, 2025
Contains on vllm-project#1111 for completeness. <!-- 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? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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? No. <!-- 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? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.com>
Angazenn
pushed a commit
to Angazenn/vllm-ascend
that referenced
this pull request
Oct 21, 2025
Contains on vllm-project#1111 for completeness. <!-- 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? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - 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? No. <!-- 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? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- 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: sdmyzlp <lrwei2@petalmail.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
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.
What this PR does / why we need it?
The current integration of
AscendW8A8DynamicLinearMethodintoCustomDeepseekV2MLPis fragile. Refactor to really utilize thequant_methodby callingapplyfunctions of various vllm-predefinedLinearBasesubclasses.Does this PR introduce any user-facing change?
No.
How was this patch tested?
Added an e2e test
test_models_distributed_DeepSeek_W8A8, usingvllm-ascend/DeepSeek-V2-Lite-W8A8for inference.