[V1][Hybrid] Mamba Prefix Caching with align mode (#30877)

Signed-off-by: huanghaoyan.hhy <huanghaoyan.hhy@alibaba-inc.com>
Signed-off-by: Chen Zhang <zhangch99@outlook.com>
Co-authored-by: Chen Zhang <zhangch99@outlook.com>
This commit is contained in:
Harry Huang
2026-01-23 09:56:48 -08:00
committed by GitHub
co-authored by Chen Zhang
parent fec9da0af4
commit 5206e5e28c
42 changed files with 1774 additions and 128 deletions
@@ -56,6 +56,7 @@ class MambaBase(AttentionLayerBase):
block_size=mamba_block_size,
page_size_padded=page_size_padded,
mamba_type=self.mamba_type,
mamba_cache_mode=vllm_config.cache_config.mamba_cache_mode,
num_speculative_blocks=(
vllm_config.speculative_config.num_speculative_tokens
if vllm_config.speculative_config