A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 08:15 UTC
[BugFix] Fix `mm_encoder_attn_backend` arg type checking (#28599)
Signed-off-by: Nick Hill <nhill@redhat.com>
Nick Hill committed
Nov 12, 2025 at 19:06 UTC
8832fff972b2f2c4a653f6cc10d3c77013b8326a
2 files changed
+6
-1
.buildkite/test-pipeline.yaml
+3
-1
@@ -57,7 +57,7 @@ steps:
57
- pytest -v -s -m 'not cpu_test' multimodal
58
- pytest -v -s utils_
59
60
-- label: Async Engine, Inputs, Utils, Worker Test (CPU) # 4 mins
60
+- label: Async Engine, Inputs, Utils, Worker, Config Test (CPU) # 4 mins
61
timeout_in_minutes: 10
62
source_file_dependencies:
63
- vllm/
@@ -66,6 +66,7 @@ steps:
66
- tests/multimodal
67
- tests/standalone_tests/lazy_imports.py
68
- tests/transformers_utils
69
+ - tests/config
70
no_gpu: true
71
commands:
72
- python3 standalone_tests/lazy_imports.py
@@ -73,6 +74,7 @@ steps:
74
- pytest -v -s test_outputs.py
75
- pytest -v -s -m 'cpu_test' multimodal
76
- pytest -v -s transformers_utils
77
+ - pytest -v -s config
78
79
- label: Python-only Installation Test # 10min
80
timeout_in_minutes: 20
vllm/config/multimodal.py
+3
@@ -170,6 +170,9 @@ class MultiModalConfig:
170
def _validate_mm_encoder_attn_backend(
171
cls, value: str | AttentionBackendEnum | None
172
) -> AttentionBackendEnum | None:
173
+ # We need to import the real type here (deferred to avoid circular import).
174
+ from vllm.attention.backends.registry import AttentionBackendEnum
175
+
176
if value is None or isinstance(value, AttentionBackendEnum):
177
return value
178