[KVPOOL]kv pool prefill save no redundancy by baxingpiaochong · Pull Request #4345 · vllm-project/vllm-ascend

baxingpiaochong · 2025-11-21T09:16:09Z

What this PR does / why we need it?

Does this PR introduce any user-facing change?

How was this patch tested?

vLLM version: v0.12.0
vLLM main: vllm-project/vllm@ad32e3e

gemini-code-assist

Code Review

This pull request introduces an optimization to avoid redundant saving of KV cache during prefill by controlling which ranks perform the save operation. The changes are mainly in mooncake_engine.py and kv_transfer.py.

My review has identified a few issues:

A critical issue in kv_transfer.py where lists passed to put_batch are handled inconsistently, which could lead to bugs.
Another critical issue in mooncake_engine.py regarding an inconsistent re-assignment of self.use_mla, which could cause incorrect behavior.
A high-severity issue in mooncake_engine.py concerning the use of a magic number, which impacts code maintainability.

I've provided specific suggestions to address these points. Overall, the direction of the PR is good, but these issues should be fixed before merging.

gemini-code-assist · 2025-11-21T09:18:03Z

+            self.m_store.put_batch(self.choose_and_shuffle(key_list),
+                                   self.choose_and_shuffle(addr_list),
+                                   self.choose_and_shuffle(size_list),
+                                   blockIds)


The blockIds list is not processed by choose_and_shuffle like the other lists, which creates an inconsistency and a potential bug. While blockIds might be unused inside put_batch currently, this is a potential source of bugs if the implementation of put_batch changes. For correctness and maintainability, all related lists should be processed in the same way.

Additionally, calling choose_and_shuffle on each list separately is inefficient. A more efficient approach would be to compute the indices to select once and apply them to all lists.

Suggested change

self.m_store.put_batch(self.choose_and_shuffle(key_list),

self.choose_and_shuffle(addr_list),

self.choose_and_shuffle(size_list),

blockIds)

self.m_store.put_batch(self.choose_and_shuffle(key_list),

self.choose_and_shuffle(addr_list),

self.choose_and_shuffle(size_list),

self.choose_and_shuffle(blockIds))

gemini-code-assist · 2025-11-21T09:18:03Z

        )

-        self.token_database = ChunkedTokenDatabase(self.metadata)
+        self.use_mla = vllm_config.model_config.is_deepseek_mla


self.use_mla is being re-assigned here. It was already initialized at lines 37-41 and used in the instantiation of MooncakeEngineMetadata at line 68. This new value is then used in the conditional at line 82. This inconsistency can lead to bugs, as self.metadata.use_mla might have a different value.

Please consolidate the logic for determining self.use_mla at the beginning of the __init__ method, before it's first used. The logic at lines 37-41 should probably be replaced by this line.

gemini-code-assist · 2025-11-21T09:18:03Z

+            if self.kv_role == "kv_producer" and self.tp_rank not in self.get_save_tp_ranks_new(
+                    sum(store_mask)):


The argument to get_save_tp_ranks_new is sum(store_mask), which is the number of tokens to store. Using this value to decide which ranks should save is a good optimization. However, sum(store_mask) is a tensor, and you should call .item() to get the Python number before passing it to get_save_tp_ranks_new.

Suggested change

if self.kv_role == "kv_producer" and self.tp_rank not in self.get_save_tp_ranks_new(

sum(store_mask)):

if self.kv_role == "kv_producer" and self.tp_rank not in self.get_save_tp_ranks_new(

store_mask.sum().item()):

gemini-code-assist · 2025-11-21T09:18:04Z

+    def get_save_tp_ranks_new(self, token_id) -> List[int]:
+        block_num = token_id.item() // 128
+        if block_num >= self.tp_size:
+            return list(range(self.tp_size))
+        else:
+            return list(range(block_num))


The method get_save_tp_ranks_new uses a magic number 128 for calculating block_num. This should be replaced with self.block_size for better readability and maintainability.
Also, the parameter name token_id is misleading as it represents the number of tokens, not a token ID. Renaming it to num_tokens would improve clarity.

Suggested change

def get_save_tp_ranks_new(self, token_id) -> List[int]:

block_num = token_id.item() // 128

if block_num >= self.tp_size:

return list(range(self.tp_size))

else:

return list(range(block_num))

def get_save_tp_ranks_new(self, num_tokens: int) -> List[int]:

block_num = num_tokens // self.block_size

if block_num >= self.tp_size:

return list(range(self.tp_size))

else:

return list(range(block_num))

github-actions · 2025-11-21T10:04:53Z

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

A PR should do only one thing, smaller PRs enable faster reviews.
Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

LCAIZJ · 2025-11-23T08:53:36Z

+        self.preferred_segment = kv_connector_extra_config.get(
+            "preferred_segment", False)
+        self.prefer_alloc_in_same_node = kv_connector_extra_config.get(
+            "prefer_alloc_in_same_node", False)


For the current Mooncake version with ADXL implementation, it is recommended to set preferred_segment = Trueand prefer_alloc_in_same_node = True. Once ADXL supports asynchronous operations in the future, these two parameters can be chosen based on the specific use case.

LCAIZJ · 2025-11-24T13:13:01Z

@wangxiyuan Could you also review this PR?

github-actions · 2025-11-28T10:12:59Z

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

github-actions · 2025-11-28T10:13:00Z

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

Signed-off-by: baxingpiaochong <771405853@qq.com>

@zzzzwwjj

I'd like to nominate @zzzzwwjj @realliujiaxu @LCAIZJ to join vLLM Ascend committer team. @zzzzwwjj --- - Review Quality‌: He has completed 80+reviews since April. 2025, include #3232 (comment), #4822 (comment), #4768 (comment) high quality review. - Sustained Contributions 15+ Valuable bug fix and refactor is very good. https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Azzzzwwjj+is%3Aclosed+review%3Aapproved Continuous optimization of code architecture https://github.com/vllm-project/vllm-ascend/pulls?q=author%3Azzzzwwjj+is%3Amerged - Quality Contribution‌: #1229 #1979 #4359 #4878 - Community Involvement‌: He lead the #1147, to refactor AscendFusedMoE at the first time. He shared topics about large-scale distributed inference and reinforcement learning on vLLM-Ascend meetup on August 2nd. @realliujiaxu --- - Review Quality‌: He has completed about [40+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3Arealliujiaxu+-author%3Arealliujiaxu+) since September, include #4868 (comment), #2275 (comment). - Sustained Contributions He has completed (17 commits)[https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged], continuously optimizing the performance of the MoE model. - Quality Contribution‌: Contributed the Flash Comm1 feature to the community, supporting both eager and aclgraph execution modes, while compatible with multiple MoE models including DeepSeek and GLM4.5. - #3334 - #3420 - #3015 co-author: - #3495 - #4868 - Community Involvement‌: 1. Completed two major refactors, enabling vllm-ascend to evolve more rapidly and robustly: [Linear module](#2867) and [rejection sampler](#4975) 2. [fixed 8 bugs](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged+bugfix+) in graph mode, spec decoding and async scheduling. @LCAIZJ --- - Review Quality‌: He's been the go-to reviewer for virtually all PD disaggregation and KV Pool related PRs, having completed [30+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3ALCAIZJ+is%3Aopen+-author%3ALCAIZJ+) since May 2025. Notable examples include [discussion_r2553887360](#4345 (comment)), [issuecomment-3540994801](#4161 (comment)), and [discussion_r2492593988](#3981 (comment)), all demonstrating thorough and insightful feedback. - Sustained and Quality Contributions: His contributions reflect a strong grasp of both ‌vLLM‌ and ‌vLLM Ascend‌ codebases, particularly in prefill-decode disaggregation and KV pool areas ([7 PRs merged](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+)). Prefill-Decode Disaggregation: Delivered KV transfer functionality using Mooncake TransferEngine and enabled layerwise KV transfer #1568 #2602 KV Pool: Developed the foundational KV Pool infrastructure and migrated it to the latest ADXL stack #2913 #3350 - Quality Contribution‌: #1568 #2602 #2913 #3350 - Community Involvement‌: He actively responds to [community issues](https://github.com/vllm-project/vllm-ascend/issues?q=is%3Aissue%20commenter%3ALCAIZJ%20is%3Aopen%20-author%3ALCAIZJ), continuously monitors functionality and accuracy issues related to PD disaggregation and KV Pool, and proactively delivers [bug fixes](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+bugfix). - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

@zzzzwwjj

…t#5152) I'd like to nominate @zzzzwwjj @realliujiaxu @LCAIZJ to join vLLM Ascend committer team. @zzzzwwjj --- - Review Quality‌: He has completed 80+reviews since April. 2025, include vllm-project#3232 (comment), vllm-project#4822 (comment), vllm-project#4768 (comment) high quality review. - Sustained Contributions 15+ Valuable bug fix and refactor is very good. https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Azzzzwwjj+is%3Aclosed+review%3Aapproved Continuous optimization of code architecture https://github.com/vllm-project/vllm-ascend/pulls?q=author%3Azzzzwwjj+is%3Amerged - Quality Contribution‌: vllm-project#1229 vllm-project#1979 vllm-project#4359 vllm-project#4878 - Community Involvement‌: He lead the vllm-project#1147, to refactor AscendFusedMoE at the first time. He shared topics about large-scale distributed inference and reinforcement learning on vLLM-Ascend meetup on August 2nd. @realliujiaxu --- - Review Quality‌: He has completed about [40+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3Arealliujiaxu+-author%3Arealliujiaxu+) since September, include vllm-project#4868 (comment), vllm-project#2275 (comment). - Sustained Contributions He has completed (17 commits)[https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged], continuously optimizing the performance of the MoE model. - Quality Contribution‌: Contributed the Flash Comm1 feature to the community, supporting both eager and aclgraph execution modes, while compatible with multiple MoE models including DeepSeek and GLM4.5. - vllm-project#3334 - vllm-project#3420 - vllm-project#3015 co-author: - vllm-project#3495 - vllm-project#4868 - Community Involvement‌: 1. Completed two major refactors, enabling vllm-ascend to evolve more rapidly and robustly: [Linear module](vllm-project#2867) and [rejection sampler](vllm-project#4975) 2. [fixed 8 bugs](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged+bugfix+) in graph mode, spec decoding and async scheduling. @LCAIZJ --- - Review Quality‌: He's been the go-to reviewer for virtually all PD disaggregation and KV Pool related PRs, having completed [30+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3ALCAIZJ+is%3Aopen+-author%3ALCAIZJ+) since May 2025. Notable examples include [discussion_r2553887360](vllm-project#4345 (comment)), [issuecomment-3540994801](vllm-project#4161 (comment)), and [discussion_r2492593988](vllm-project#3981 (comment)), all demonstrating thorough and insightful feedback. - Sustained and Quality Contributions: His contributions reflect a strong grasp of both ‌vLLM‌ and ‌vLLM Ascend‌ codebases, particularly in prefill-decode disaggregation and KV pool areas ([7 PRs merged](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+)). Prefill-Decode Disaggregation: Delivered KV transfer functionality using Mooncake TransferEngine and enabled layerwise KV transfer vllm-project#1568 vllm-project#2602 KV Pool: Developed the foundational KV Pool infrastructure and migrated it to the latest ADXL stack vllm-project#2913 vllm-project#3350 - Quality Contribution‌: vllm-project#1568 vllm-project#2602 vllm-project#2913 vllm-project#3350 - Community Involvement‌: He actively responds to [community issues](https://github.com/vllm-project/vllm-ascend/issues?q=is%3Aissue%20commenter%3ALCAIZJ%20is%3Aopen%20-author%3ALCAIZJ), continuously monitors functionality and accuracy issues related to PD disaggregation and KV Pool, and proactively delivers [bug fixes](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+bugfix). - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

wangxiyuan · 2026-01-05T06:33:02Z

Any progress? If this PR is still alive, please rebase to main and make CI happy, otherwise you can close it. Thanks

@zzzzwwjj

…t#5152) I'd like to nominate @zzzzwwjj @realliujiaxu @LCAIZJ to join vLLM Ascend committer team. @zzzzwwjj --- - Review Quality‌: He has completed 80+reviews since April. 2025, include vllm-project#3232 (comment), vllm-project#4822 (comment), vllm-project#4768 (comment) high quality review. - Sustained Contributions 15+ Valuable bug fix and refactor is very good. https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Azzzzwwjj+is%3Aclosed+review%3Aapproved Continuous optimization of code architecture https://github.com/vllm-project/vllm-ascend/pulls?q=author%3Azzzzwwjj+is%3Amerged - Quality Contribution‌: vllm-project#1229 vllm-project#1979 vllm-project#4359 vllm-project#4878 - Community Involvement‌: He lead the vllm-project#1147, to refactor AscendFusedMoE at the first time. He shared topics about large-scale distributed inference and reinforcement learning on vLLM-Ascend meetup on August 2nd. @realliujiaxu --- - Review Quality‌: He has completed about [40+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3Arealliujiaxu+-author%3Arealliujiaxu+) since September, include vllm-project#4868 (comment), vllm-project#2275 (comment). - Sustained Contributions He has completed (17 commits)[https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged], continuously optimizing the performance of the MoE model. - Quality Contribution‌: Contributed the Flash Comm1 feature to the community, supporting both eager and aclgraph execution modes, while compatible with multiple MoE models including DeepSeek and GLM4.5. - vllm-project#3334 - vllm-project#3420 - vllm-project#3015 co-author: - vllm-project#3495 - vllm-project#4868 - Community Involvement‌: 1. Completed two major refactors, enabling vllm-ascend to evolve more rapidly and robustly: [Linear module](vllm-project#2867) and [rejection sampler](vllm-project#4975) 2. [fixed 8 bugs](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged+bugfix+) in graph mode, spec decoding and async scheduling. @LCAIZJ --- - Review Quality‌: He's been the go-to reviewer for virtually all PD disaggregation and KV Pool related PRs, having completed [30+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3ALCAIZJ+is%3Aopen+-author%3ALCAIZJ+) since May 2025. Notable examples include [discussion_r2553887360](vllm-project#4345 (comment)), [issuecomment-3540994801](vllm-project#4161 (comment)), and [discussion_r2492593988](vllm-project#3981 (comment)), all demonstrating thorough and insightful feedback. - Sustained and Quality Contributions: His contributions reflect a strong grasp of both ‌vLLM‌ and ‌vLLM Ascend‌ codebases, particularly in prefill-decode disaggregation and KV pool areas ([7 PRs merged](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+)). Prefill-Decode Disaggregation: Delivered KV transfer functionality using Mooncake TransferEngine and enabled layerwise KV transfer vllm-project#1568 vllm-project#2602 KV Pool: Developed the foundational KV Pool infrastructure and migrated it to the latest ADXL stack vllm-project#2913 vllm-project#3350 - Quality Contribution‌: vllm-project#1568 vllm-project#2602 vllm-project#2913 vllm-project#3350 - Community Involvement‌: He actively responds to [community issues](https://github.com/vllm-project/vllm-ascend/issues?q=is%3Aissue%20commenter%3ALCAIZJ%20is%3Aopen%20-author%3ALCAIZJ), continuously monitors functionality and accuracy issues related to PD disaggregation and KV Pool, and proactively delivers [bug fixes](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+bugfix). - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> Signed-off-by: zrj026 <zhangrunjiang026@gmail.com>

@zzzzwwjj

…t#5152) I'd like to nominate @zzzzwwjj @realliujiaxu @LCAIZJ to join vLLM Ascend committer team. @zzzzwwjj --- - Review Quality‌: He has completed 80+reviews since April. 2025, include vllm-project#3232 (comment), vllm-project#4822 (comment), vllm-project#4768 (comment) high quality review. - Sustained Contributions 15+ Valuable bug fix and refactor is very good. https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Azzzzwwjj+is%3Aclosed+review%3Aapproved Continuous optimization of code architecture https://github.com/vllm-project/vllm-ascend/pulls?q=author%3Azzzzwwjj+is%3Amerged - Quality Contribution‌: vllm-project#1229 vllm-project#1979 vllm-project#4359 vllm-project#4878 - Community Involvement‌: He lead the vllm-project#1147, to refactor AscendFusedMoE at the first time. He shared topics about large-scale distributed inference and reinforcement learning on vLLM-Ascend meetup on August 2nd. @realliujiaxu --- - Review Quality‌: He has completed about [40+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3Arealliujiaxu+-author%3Arealliujiaxu+) since September, include vllm-project#4868 (comment), vllm-project#2275 (comment). - Sustained Contributions He has completed (17 commits)[https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged], continuously optimizing the performance of the MoE model. - Quality Contribution‌: Contributed the Flash Comm1 feature to the community, supporting both eager and aclgraph execution modes, while compatible with multiple MoE models including DeepSeek and GLM4.5. - vllm-project#3334 - vllm-project#3420 - vllm-project#3015 co-author: - vllm-project#3495 - vllm-project#4868 - Community Involvement‌: 1. Completed two major refactors, enabling vllm-ascend to evolve more rapidly and robustly: [Linear module](vllm-project#2867) and [rejection sampler](vllm-project#4975) 2. [fixed 8 bugs](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3Arealliujiaxu+is%3Amerged+bugfix+) in graph mode, spec decoding and async scheduling. @LCAIZJ --- - Review Quality‌: He's been the go-to reviewer for virtually all PD disaggregation and KV Pool related PRs, having completed [30+ reviews](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+commenter%3ALCAIZJ+is%3Aopen+-author%3ALCAIZJ+) since May 2025. Notable examples include [discussion_r2553887360](vllm-project#4345 (comment)), [issuecomment-3540994801](vllm-project#4161 (comment)), and [discussion_r2492593988](vllm-project#3981 (comment)), all demonstrating thorough and insightful feedback. - Sustained and Quality Contributions: His contributions reflect a strong grasp of both ‌vLLM‌ and ‌vLLM Ascend‌ codebases, particularly in prefill-decode disaggregation and KV pool areas ([7 PRs merged](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+)). Prefill-Decode Disaggregation: Delivered KV transfer functionality using Mooncake TransferEngine and enabled layerwise KV transfer vllm-project#1568 vllm-project#2602 KV Pool: Developed the foundational KV Pool infrastructure and migrated it to the latest ADXL stack vllm-project#2913 vllm-project#3350 - Quality Contribution‌: vllm-project#1568 vllm-project#2602 vllm-project#2913 vllm-project#3350 - Community Involvement‌: He actively responds to [community issues](https://github.com/vllm-project/vllm-ascend/issues?q=is%3Aissue%20commenter%3ALCAIZJ%20is%3Aopen%20-author%3ALCAIZJ), continuously monitors functionality and accuracy issues related to PD disaggregation and KV Pool, and proactively delivers [bug fixes](https://github.com/vllm-project/vllm-ascend/pulls?q=is%3Apr+author%3ALCAIZJ+is%3Amerged+bugfix). - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com> Signed-off-by: zrj026 <zhangrunjiang026@gmail.com>

gemini-code-assist bot reviewed Nov 21, 2025

View reviewed changes

LCAIZJ reviewed Nov 23, 2025

View reviewed changes

LCAIZJ approved these changes Nov 24, 2025

View reviewed changes

github-actions bot added the merge-conflicts label Nov 28, 2025

kv pool prefill save no redundancy

b37eac4

Signed-off-by: baxingpiaochong <771405853@qq.com>

baxingpiaochong force-pushed the no_duplicate_save_new branch from a2a4ed8 to b37eac4 Compare December 6, 2025 08:58

wangxiyuan mentioned this pull request Dec 18, 2025

Nominate new maintainers @zzzzwwjj @realliujiaxu @LCAIZJ #5152

Merged

baxingpiaochong closed this Jan 5, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[KVPOOL]kv pool prefill save no redundancy#4345

[KVPOOL]kv pool prefill save no redundancy#4345
baxingpiaochong wants to merge 1 commit intovllm-project:mainfrom
baxingpiaochong:no_duplicate_save_new

baxingpiaochong commented Nov 21, 2025 •

edited by github-actions bot

Loading

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Nov 21, 2025

Uh oh!

gemini-code-assist bot Nov 21, 2025

Uh oh!

gemini-code-assist bot Nov 21, 2025

Uh oh!

gemini-code-assist bot Nov 21, 2025

Uh oh!

github-actions bot commented Nov 21, 2025

Uh oh!

LCAIZJ Nov 23, 2025

Uh oh!

LCAIZJ commented Nov 24, 2025

Uh oh!

github-actions bot commented Nov 28, 2025

Uh oh!

github-actions bot commented Nov 28, 2025

Uh oh!

wangxiyuan commented Jan 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		if self.kv_role == "kv_producer" and self.tp_rank not in self.get_save_tp_ranks_new(
		sum(store_mask)):

Conversation

baxingpiaochong commented Nov 21, 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!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Nov 21, 2025

Uh oh!

LCAIZJ Nov 23, 2025

Choose a reason for hiding this comment

Uh oh!

LCAIZJ commented Nov 24, 2025

Uh oh!

github-actions bot commented Nov 28, 2025

Uh oh!

github-actions bot commented Nov 28, 2025

Uh oh!

wangxiyuan commented Jan 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

baxingpiaochong commented Nov 21, 2025 •

edited by github-actions bot

Loading