Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
vllm_cscc
Commits
b51255f3
Unverified
Commit
b51255f3
authored
Dec 10, 2025
by
Andreas Karatzas
Committed by
GitHub
Dec 11, 2025
Browse files
[ROCm] Fix broken import in platform attention backend dispatching (#30432)
Signed-off-by:
Andreas Karatzas
<
akaratza@amd.com
>
parent
b4054c8a
Changes
1
Show whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
15 additions
and
1 deletion
+15
-1
vllm/platforms/rocm.py
vllm/platforms/rocm.py
+15
-1
No files found.
vllm/platforms/rocm.py
View file @
b51255f3
...
...
@@ -403,6 +403,20 @@ class RocmPlatform(Platform):
compilation_config
.
cudagraph_mode
=
CUDAGraphMode
.
PIECEWISE
if
cache_config
and
cache_config
.
block_size
is
None
:
if
(
envs
.
VLLM_ROCM_USE_AITER_UNIFIED_ATTENTION
and
envs
.
VLLM_ROCM_USE_AITER
# NOTE: This block has been deprecated
# or get_env_variable_attn_backend()
# == AttentionBackendEnum.ROCM_AITER_UNIFIED_ATTN
# TODO: monitor https://github.com/vllm-project/vllm/pull/30396
# to see how we can transition to the new way of selecting
# attention backends
):
cache_config
.
block_size
=
64
logger
.
warning
(
"[ROCM_AITER_UNIFIED_ATTN]: Setting kv cache block size to 64."
)
else
:
cache_config
.
block_size
=
16
if
parallel_config
.
worker_cls
==
"auto"
:
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment