[KV Connector] Support symmetric DCP disagg for hybrid mamba models (#55531)

Signed-off-by: Wei Zhao <weizha@oci-aga-slurm-1-vscode-02.cm.cluster>
Signed-off-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Co-authored-by: Wei Zhao <weizha@oci-aga-slurm-1-vscode-02.cm.cluster>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Wei Zhao
2026-09-10 11:12:15 -07:00
committed by GitHub
co-authored by Wei Zhao OpenAI Codex mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 983b7e28c9
commit 7cdd9304ae
5 changed files with 31 additions and 12 deletions
+7
View File
@@ -393,8 +393,15 @@ class ParallelConfig:
in (rank i+1, block j) only after (rank i, block j) is fully occupied.
Block_size should be greater than or equal to cp_kv_cache_interleave_size.
Block_size should be divisible by cp_kv_cache_interleave_size.
When --cp-kv-cache-interleave-size is omitted (None), the interleave size
is resolved automatically based on NIXL transfer requirements.
Explicit settings take priority.
"""
_allow_auto_resolve_cp_interleave_size: bool = True
"""Whether NIXL may select the interleave size automatically."""
data_parallel_index: int = Field(init=False)
"""Equal to the data parallel rank but not used for torch process groups
and not overridden for dense models."""