[Model Runner V2] support qwen35 / mamba hybrid model (#35520)

Signed-off-by: zhuhaoran <zhuhaoran.zhr@alibaba-inc.com>
This commit is contained in:
zhrrr
2026-05-07 09:31:05 -07:00
committed by GitHub
parent 06a60d3dd0
commit 7a08b34fbf
14 changed files with 439 additions and 106 deletions
@@ -106,6 +106,7 @@ class DecodeEagleCudaGraphManager(EagleCudaGraphManagerBase):
block_tables,
attn_groups,
kv_cache_config,
skip_attn=(desc.cg_mode == CUDAGraphMode.PIECEWISE),
)
attn_metadata, slot_mappings = attn_state