xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-06 16:49:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Correct num_q_heads on DCP for Flashinfer backends (#29487)

Browse Source 
Signed-off-by: Jingchun Gao <gaojingchun1@huawei.com>
Signed-off-by: Jingchun Gao <63247409+gjc0824@users.noreply.github.com>
Co-authored-by: Jingchun Gao <gaojingchun1@huawei.com>
This commit is contained in:
Jingchun Gao 2025-12-05 13:54:31 +08:00 committed by GitHub
parent 2c22c4ca2d
commit d698bb382d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/v1/attention/backends/flashinfer.py
View File

@ -482,9 +482,8 @@ class FlashInferMetadataBuilder(AttentionMetadataBuilder[FlashInferMetadata]):
self.dcp_rank = 0 self.dcp_rank = 0
self.dcp_kv_cache_interleave_size = 1 self.dcp_kv_cache_interleave_size = 1
self.num_qo_heads = ( self.num_qo_heads = self.model_config.get_num_attention_heads(
self.model_config.get_num_attention_heads(self.vllm_config.parallel_config) self.vllm_config.parallel_config
* self.dcp_world_size
) )
self.num_kv_heads = self.kv_cache_spec.num_kv_heads self.num_kv_heads = self.kv_cache_spec.num_kv_heads