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

fix some typos (#24071)

Browse Source 
Signed-off-by: co63oc <co63oc@users.noreply.github.com>
This commit is contained in:
co63oc
2025-09-03 11:44:50 +08:00
committed by GitHub
parent 136d853e65
commit 1bd007f234
32 changed files with 39 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -167,7 +167,7 @@ class FlashAttentionMetadataBuilder(
# work for mixed prefill-decode and uniform-decode. But for non-spec decodes
# the graphs would not work for mixed prefill-decode; sorta the inverse
# of UNIFORM_SINGLE_TOKEN_DECODE.
# Theres probably a better way to describe this using `AttentionCGSupport`
# There's probably a better way to describe this using `AttentionCGSupport`
# but for now just set it to `UNIFORM_BATCH` to get use to drop down
# to FULL_AND_PIECEWISE.
# TODO(luka, lucas): audit FA2 as part of:

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

@@ -291,7 +291,7 @@ class FlashInferMetadataBuilder(AttentionMetadataBuilder[FlashInferMetadata]):
paged_kv_indices_buffer=paged_kv_indices,
paged_kv_last_page_len_buffer=paged_kv_last_page_len,
# Tensor cores are enabled by default because the perf would be
# atleast as good as cuda cores for all attention ops in latest
# at least as good as cuda cores for all attention ops in latest
# gpus.
use_tensor_cores=True,
)