From be292b7c14e08e6e6883d5ebee79240d04814159 Mon Sep 17 00:00:00 2001 From: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Date: Mon, 9 Mar 2026 11:17:45 -0400 Subject: [PATCH] [Bug] Fix pooling model benchmark script (#36300) Signed-off-by: yewentao256 --- vllm/benchmarks/lib/endpoint_request_func.py | 11 +++++++++++ vllm/benchmarks/serve.py | 7 ++----- 2 files changed, 13 insertions(+), 5 deletions(-) diff --git a/vllm/benchmarks/lib/endpoint_request_func.py b/vllm/benchmarks/lib/endpoint_request_func.py index e231ccf6e1..b0ef67889d 100644 --- a/vllm/benchmarks/lib/endpoint_request_func.py +++ b/vllm/benchmarks/lib/endpoint_request_func.py @@ -795,6 +795,17 @@ ASYNC_REQUEST_FUNCS: dict[str, RequestFunc] = { "vllm-rerank": async_request_vllm_rerank, } +POOLING_BACKENDS = { + "openai-embeddings", + "openai-embeddings-chat", + "openai-embeddings-clip", + "openai-embeddings-vlm2vec", + "infinity-embeddings", + "infinity-embeddings-clip", + "vllm-pooling", + "vllm-rerank", +} + OPENAI_COMPATIBLE_BACKENDS = [ k for k, v in ASYNC_REQUEST_FUNCS.items() diff --git a/vllm/benchmarks/serve.py b/vllm/benchmarks/serve.py index 7c9a95ef12..fca01e17ea 100644 --- a/vllm/benchmarks/serve.py +++ b/vllm/benchmarks/serve.py @@ -45,6 +45,7 @@ from vllm.benchmarks.datasets import SampleRequest, add_dataset_parser, get_samp from vllm.benchmarks.lib.endpoint_request_func import ( ASYNC_REQUEST_FUNCS, OPENAI_COMPATIBLE_BACKENDS, + POOLING_BACKENDS, RequestFuncInput, RequestFuncOutput, ) @@ -1721,11 +1722,7 @@ async def main_async(args: argparse.Namespace) -> dict[str, Any]: goodput_config_dict = check_goodput_args(args) backend = args.backend - task_type = ( - TaskType.POOLING - if "embeddings" in backend or "rerank" in backend - else TaskType.GENERATION - ) + task_type = TaskType.POOLING if backend in POOLING_BACKENDS else TaskType.GENERATION # Collect the sampling parameters. if task_type == TaskType.GENERATION: