biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[backends][short_conv] CUDA graph piecewise edits (#24215)

Browse Source 
Signed-off-by: Paul Pak <paulpak58@gmail.com>
This commit is contained in:
Paul Pak
2025-10-03 21:59:48 +09:00
committed by GitHub
parent 8ee846c27c
commit 5f42fc53b6
2 changed files with 21 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/layers/mamba/short_conv.py
View File

@@ -115,7 +115,7 @@ class ShortConv(MambaBase, CustomOp):
self_kv_cache = self.kv_cache[forward_context.virtual_engine]
conv_state = self_kv_cache[0].transpose(-1, -2)
state_indices_tensor = attn_metadata.state_indices_tensor
has_initial_states_p = attn_metadata.has_initial_states
has_initial_states_p = attn_metadata.has_initial_states_p
BCx, _ = self.in_proj(hidden_states)