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

[Misc] Delay deprecation of CommonAttentionMetadata properties (#33801)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Lucas Wilkinson
2026-02-04 09:41:57 -07:00
committed by GitHub
parent 87d9a26166
commit 0e92298622
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/v1/attention/backend.py
View File

@@ -347,7 +347,7 @@ class CommonAttentionMetadata:
"""
Prefer using device seq_lens directly to avoid implicit H<>D sync.
If a CPU copy is needed, use `seq_lens.cpu()` instead.
Will be removed in a future release (v0.15.0)
Will be removed in a future release, please migrate as soon as possible.
"""
)
def seq_lens_cpu(self) -> torch.Tensor:
@@ -361,7 +361,7 @@ class CommonAttentionMetadata:
Prefer using device seq_lens directly to avoid implicit H<>D sync which breaks full
async scheduling. If a CPU copy is needed, it can be derived from
query_start_loc_cpu and seq_lens.
Will be removed in a future release (v0.15.0)
Will be removed in a future release, please migrate as soon as possible.
"""
)
def num_computed_tokens_cpu(self) -> torch.Tensor: