A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 00:15 UTC
[BugFix] Remove incorrect assert in split_decodes_and_prefills (#36553)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Woosuk Kwon committed
Mar 9, 2026 at 20:02 UTC
006aea17d7de338ab9f9e13bfe566715782d19a4
1 file changed
-1
vllm/v1/attention/backends/utils.py
-1
@@ -528,7 +528,6 @@ def split_decodes_and_prefills(
528
# requests may have a query length of 0 but since they are padding its fine
529
# to treat them as decodes (ensures num_decodes matches the captured size)
530
if torch.all((query_lens == query_lens[0]) | (query_lens == 0)):
531
- assert num_reqs * query_lens[0] == num_tokens, "tokens not padded correctly"
531
return num_reqs, 0, num_tokens, 0 # all decodes
532
is_prefill = query_lens != query_lens[0]
533
else: