[bugfix] Fix MHA model runtime error in aclgraph mode - #5397
Merged
Merged
Conversation
![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 aims to fix a runtime error for MHA models in aclgraph mode. It introduces a change in the attention mechanism to correctly slice key and value tensors during the prefill stage. Additionally, it expands test coverage by parameterizing an existing test to include both MHA and GQA models. My review identifies a critical issue where the query tensor is not being sliced consistently with the key and value tensors in the aclgraph capture path, which could lead to shape mismatch errors. A fix is suggested to ensure tensor shapes are consistent.
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. |
weijinqian0
approved these changes
Dec 26, 2025
weijinqian0
reviewed
Dec 26, 2025
| actual_seq_lengths_kv = attn_metadata.actual_seq_lengths_q | ||
| num_tokens_kv = actual_seq_lengths_kv[-1] | ||
| key = key[:num_tokens_kv] | ||
| value = value[:num_tokens_kv] |
Collaborator
There was a problem hiding this comment.
move this to eager or piecewise branch.
weijinqian0
self-requested a review
December 26, 2025 09:45
46 tasks
Signed-off-by: Wang Kunpeng <1289706727@qq.com>
kunpengW-code
force-pushed
the
bugfix-mha
branch
from
December 26, 2025 09:59
26582e0 to
51c3f5c
Compare
weijinqian0
approved these changes
Dec 26, 2025
maoxx241
pushed a commit
to maoxx241/vllm-ascend
that referenced
this pull request
Mar 2, 2026
) ### What this PR does / why we need it? Currently, MHA models (eg: minicpm-2b, Baichuan-7b) will encounter errors when running in piecewise graph mode, with error messages similar to: ``` (E89999): When layout is TND and PA not enabled, keyT(8) and valueT(8) must be equal to the last element of actualSeqenceLengthKV(5)[FUNC:CheckInputShapeWhenLayoutIsTND][FILE:prompt_flash_attention_tiling.cpp][LINE:3618] ``` The error occurs because the qkv in the Prefill stage is also padded, causing the shape to be inconsistent with actual_seq_lengths. Add unpadding logic for kv. - vLLM version: release/v0.13.0 - vLLM main: vllm-project/vllm@254f6b9 Signed-off-by: Wang Kunpeng <1289706727@qq.com>
yangzhe-2026
pushed a commit
to yangzhe-2026/vllm-ascend
that referenced
this pull request
May 6, 2026
) ### What this PR does / why we need it? Currently, MHA models (eg: minicpm-2b, Baichuan-7b) will encounter errors when running in piecewise graph mode, with error messages similar to: ``` (E89999): When layout is TND and PA not enabled, keyT(8) and valueT(8) must be equal to the last element of actualSeqenceLengthKV(5)[FUNC:CheckInputShapeWhenLayoutIsTND][FILE:prompt_flash_attention_tiling.cpp][LINE:3618] ``` The error occurs because the qkv in the Prefill stage is also padded, causing the shape to be inconsistent with actual_seq_lengths. Add unpadding logic for kv. - vLLM version: release/v0.13.0 - vLLM main: vllm-project/vllm@254f6b9 Signed-off-by: Wang Kunpeng <1289706727@qq.com>
nanxingMy
pushed a commit
to nanxingMy/vllm-ascend
that referenced
this pull request
May 15, 2026
) ### What this PR does / why we need it? Currently, MHA models (eg: minicpm-2b, Baichuan-7b) will encounter errors when running in piecewise graph mode, with error messages similar to: ``` (E89999): When layout is TND and PA not enabled, keyT(8) and valueT(8) must be equal to the last element of actualSeqenceLengthKV(5)[FUNC:CheckInputShapeWhenLayoutIsTND][FILE:prompt_flash_attention_tiling.cpp][LINE:3618] ``` The error occurs because the qkv in the Prefill stage is also padded, causing the shape to be inconsistent with actual_seq_lengths. Add unpadding logic for kv. - vLLM version: release/v0.13.0 - vLLM main: vllm-project/vllm@254f6b9 Signed-off-by: Wang Kunpeng <1289706727@qq.com> Signed-off-by: nanxing <1014662416@qq.com>
ader47
pushed a commit
to ader47/vllm-ascend
that referenced
this pull request
Jun 18, 2026
) ### What this PR does / why we need it? Currently, MHA models (eg: minicpm-2b, Baichuan-7b) will encounter errors when running in piecewise graph mode, with error messages similar to: ``` (E89999): When layout is TND and PA not enabled, keyT(8) and valueT(8) must be equal to the last element of actualSeqenceLengthKV(5)[FUNC:CheckInputShapeWhenLayoutIsTND][FILE:prompt_flash_attention_tiling.cpp][LINE:3618] ``` The error occurs because the qkv in the Prefill stage is also padded, causing the shape to be inconsistent with actual_seq_lengths. Add unpadding logic for kv. - vLLM version: release/v0.13.0 - vLLM main: vllm-project/vllm@254f6b9 Signed-off-by: Wang Kunpeng <1289706727@qq.com>
CXY-Katrina
pushed a commit
to CXY-Katrina/vllm-ascend
that referenced
this pull request
Jun 27, 2026
) ### What this PR does / why we need it? Currently, MHA models (eg: minicpm-2b, Baichuan-7b) will encounter errors when running in piecewise graph mode, with error messages similar to: ``` (E89999): When layout is TND and PA not enabled, keyT(8) and valueT(8) must be equal to the last element of actualSeqenceLengthKV(5)[FUNC:CheckInputShapeWhenLayoutIsTND][FILE:prompt_flash_attention_tiling.cpp][LINE:3618] ``` The error occurs because the qkv in the Prefill stage is also padded, causing the shape to be inconsistent with actual_seq_lengths. Add unpadding logic for kv. - vLLM version: release/v0.13.0 - vLLM main: vllm-project/vllm@254f6b9 Signed-off-by: Wang Kunpeng <1289706727@qq.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
4 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?
Currently, MHA models (eg: minicpm-2b, Baichuan-7b) will encounter errors when running in piecewise graph mode, with error messages similar to:
The error occurs because the qkv in the Prefill stage is also padded, causing the shape to be inconsistent with actual_seq_lengths.
Add unpadding logic for kv.
Does this PR introduce any user-facing change?
no
How was this patch tested?