biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix] Fix AssertionError: DCP not support reorder_batch_threshold > 1 now. (#28751)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Lucas Wilkinson
2025-11-15 17:35:06 -05:00
committed by GitHub
parent 2bb4435cb7
commit be263f7645
1 changed files with 0 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/v1/worker/gpu_model_runner.py
View File

@@ -630,16 +630,6 @@ class GPUModelRunner(
return
if self.reorder_batch_threshold is not None:
# NOTE(lucas): currently no backend supports the custom masking
# required for DCP with q_len > 1, so we assert here. Remove this
# assert once the custom mask is support is added to FA3.
if (
self.dcp_world_size > 1
and envs.VLLM_ATTENTION_BACKEND != "FLASH_ATTN_MLA"
):
assert self.reorder_batch_threshold == 1, (
"DCP not support reorder_batch_threshold > 1 now."
)
reorder_batch_to_split_decodes_and_prefills(
self.input_batch,
scheduler_output,