[DCP] Default query replication for GLM sparse attention (#50382)

Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Lucas Wilkinson
2026-08-21 09:07:05 -07:00
committed by GitHub
co-authored by OpenAI Codex Claude Opus 5
parent cd7b7c265a
commit 72aedcc426
6 changed files with 62 additions and 20 deletions
+7 -1
View File
@@ -481,7 +481,8 @@ class EngineArgs:
tensor_parallel_size: int = ParallelConfig.tensor_parallel_size
prefill_context_parallel_size: int = ParallelConfig.prefill_context_parallel_size
decode_context_parallel_size: int = ParallelConfig.decode_context_parallel_size
dcp_comm_backend: DCPCommBackend = ParallelConfig.dcp_comm_backend
dcp_comm_backend: DCPCommBackend | None = ParallelConfig.dcp_comm_backend
dcp_q_replicate: bool | None = ParallelConfig.dcp_q_replicate
dcp_kv_cache_interleave_size: int = ParallelConfig.dcp_kv_cache_interleave_size
cp_kv_cache_interleave_size: int = ParallelConfig.cp_kv_cache_interleave_size
data_parallel_size: int = ParallelConfig.data_parallel_size
@@ -1070,6 +1071,10 @@ class EngineArgs:
"--dcp-comm-backend",
**parallel_kwargs["dcp_comm_backend"],
)
parallel_group.add_argument(
"--dcp-q-replicate",
**parallel_kwargs["dcp_q_replicate"],
)
parallel_group.add_argument(
"--dcp-kv-cache-interleave-size",
**parallel_kwargs["dcp_kv_cache_interleave_size"],
@@ -2295,6 +2300,7 @@ class EngineArgs:
worker_extension_cls=self.worker_extension_cls,
decode_context_parallel_size=self.decode_context_parallel_size,
dcp_comm_backend=self.dcp_comm_backend,
dcp_q_replicate=self.dcp_q_replicate,
dcp_kv_cache_interleave_size=self.dcp_kv_cache_interleave_size,
cp_kv_cache_interleave_size=self.cp_kv_cache_interleave_size,
_api_process_count=self._api_process_count,