[V0 Deprecation] Refactor kv cache from list to element (#37487)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
2026-03-23 23:10:11 -04:00
parent de99d91ece
commit c59a132f96
27 changed files with 70 additions and 85 deletions
--- a/vllm/model_executor/layers/mamba/mamba_mixer2.py
+++ b/vllm/model_executor/layers/mamba/mamba_mixer2.py
@@ -575,7 +575,7 @@ class MambaMixer2(MambaBase, PluggableLayer):
            assert isinstance(attn_metadata, dict)
            attn_metadata = attn_metadata[self.prefix]
            assert isinstance(attn_metadata, Mamba2AttentionMetadata)
-            self_kv_cache = self.kv_cache[0]
+            self_kv_cache = self.kv_cache
            # conv_state = (..., dim, width-1) yet contiguous along 'dim'
            conv_state = self_kv_cache[0].transpose(-1, -2)
            ssm_state = self_kv_cache[1]