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

[Attention][1/n] Remove usage of deprecated seq_lens_cpu and num_computed_tokens_cpu CommonAttentionMetadata properties (#31773)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Lucas Wilkinson
2026-01-06 07:05:17 -05:00
committed by GitHub
parent 14df02b4e1
commit e0327c9db2
9 changed files with 23 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -221,7 +221,7 @@ class TritonAttentionMetadataBuilder(AttentionMetadataBuilder[TritonAttentionMet
prefix_kv_lens = torch.tensor(
[common_prefix_len], dtype=torch.int32, device=self.device
)
suffix_kv_lens = common_attn_metadata.seq_lens_cpu - common_prefix_len
suffix_kv_lens = common_attn_metadata.seq_lens.cpu() - common_prefix_len
suffix_kv_lens = suffix_kv_lens.to(self.device)
else:
cu_prefix_query_lens = None