[ModelRunner V2] Enable by default for all dense models (#44443)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
Wentao Ye
2026-07-02 18:48:57 +08:00
committed by GitHub
co-authored by Nick Hill
parent de2a8fc042
commit a2f713002d
6 changed files with 55 additions and 17 deletions
+8 -3
View File
@@ -283,9 +283,14 @@ def test_multiproc_executor_pipeline_parallel():
output_rank = executor._get_output_rank()
assert output_rank == 2, "Output rank should be 2 (first rank of last PP stage)"
# Verify max_concurrent_batches for pipeline parallel
assert vllm_config.max_concurrent_batches == 2, (
"Max concurrent batches should equal PP size"
# V2 model runner uses one extra batch to overlap async scheduling.
expected_concurrent_batches = 2 + int(
vllm_config.scheduler_config.async_scheduling
and vllm_config.use_v2_model_runner
)
assert vllm_config.max_concurrent_batches == expected_concurrent_batches, (
"Max concurrent batches should follow the configured PP/async "
"scheduling policy"
)
finally: