[Attention] Remove DCP indexer interleave guard and test TP1 output parity (#55802)
Signed-off-by: Stu Cao <stucao@Stus-MacBook-Pro.local> Signed-off-by: <> Signed-off-by: Summer Yang <girasoleyang@gmail.com> Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: Stu Cao <stucao@Stus-MacBook-Pro.local> Co-authored-by: OpenAI Codex <codex@openai.com> Co-authored-by: Summer Yang <girasoleyang@gmail.com> Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: OpenAI Codex <noreply@openai.com>
stu-cao committed
Sep 13, 2026 at 13:56 UTC
b23433088bf29980d20dff1bd64c753dd8883506
1 file changed
-9
vllm/v1/attention/backends/mla/indexer.py
-9
@@ -771,15 +771,6 @@ class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder):
771
self.pcp_world_size = parallel_config.prefill_context_parallel_size
772
self.use_pcp = self.pcp_world_size > 1
773
self.cp_kv_cache_interleave_size = parallel_config.cp_kv_cache_interleave_size
774
- # The DCP sparse-indexer code is parameterized by interleave size, but
775
- # interleave > 1 is not yet validated end-to-end (gsm8k parity fails),
776
- # so fail closed here rather than silently produce wrong output.
777
- if self.dcp_world_size > 1 and self.cp_kv_cache_interleave_size > 1:
778
- raise NotImplementedError(
779
- "DCP sparse indexer currently supports only "
780
- f"cp_kv_cache_interleave_size=1 (got "
781
- f"{self.cp_kv_cache_interleave_size})."
782
- )
774
# NOTE(Chen):an estimated max size of flattened_kv. Need to double check.
775
self.max_prefill_buffer_size = get_max_prefill_buffer_size(self.vllm_config)
776
self.num_speculative_tokens = (