xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-17 06:57:07 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix][ROCm] Fix uninitialized prefix_scheduler_metadata variable

Browse Source 
When use_cascade=True, the prefix_scheduler_metadata variable was only
initialized in the else branch, causing an UnboundLocalError when
use_cascade was True.

Move initialization before the if/else block to ensure the variable
is always defined.

Signed-off-by: c0de128 <kevin.mckay@outlook.com>
This commit is contained in:
c0de128 2025-12-21 22:50:58 -06:00
parent 19cc9468fd
commit e85f4a3686
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/attention/backends/rocm_attn.py
View File

@ -116,6 +116,7 @@ class RocmAttentionMetadataBuilder(AttentionMetadataBuilder[RocmAttentionMetadat
slot_mapping = common_attn_metadata.slot_mapping
use_cascade = common_prefix_len > 0
prefix_scheduler_metadata = None
if use_cascade:
cu_prefix_query_lens = torch.tensor(
@ -130,7 +131,6 @@ class RocmAttentionMetadataBuilder(AttentionMetadataBuilder[RocmAttentionMetadat
cu_prefix_query_lens = None
prefix_kv_lens = None
suffix_kv_lens = None
prefix_scheduler_metadata = None
attn_metadata = RocmAttentionMetadata(
num_actual_tokens=num_actual_tokens,