karylab_agents/vllm
1
0
Fork 0
forked from Karylab-cklius/vllm
Code Pull Requests 2 Actions 2 Packages Activity

Update vllm/v1/worker/gpu/model_states/default.py

Browse Source 
Co-authored-by: Nick Hill <nhill@redhat.com>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Wentao Ye
2026-03-04 13:57:18 -05:00
committed by GitHub
co-authored by Nick Hill
parent b49f48c46b
commit e24a1df570
1 changed files with 3 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/v1/worker/gpu/model_states/default.py
+3 -8
View File
@@ -195,15 +195,10 @@ class DefaultModelState(ModelState):
and num_reqs > 0
and bool((query_lens_np == self.uniform_decode_query_len).all())
)
attn_num_reqs = input_batch.num_tokens_after_padding
if is_uniform_full_decode:
attn_num_reqs = min(
cdiv(
input_batch.num_tokens_after_padding, self.uniform_decode_query_len
),
self.max_num_reqs,
)
else:
attn_num_reqs = min(input_batch.num_tokens_after_padding, self.max_num_reqs)
attn_num_reqs = cdiv(attn_num_reqs, self.uniform_decode_query_len)
attn_num_reqs = min(attn_num_reqs, self.max_num_reqs)
attn_num_tokens = input_batch.num_tokens_after_padding
attn_query_start_loc_cpu = torch.empty(attn_num_reqs + 1, dtype=torch.int32)