From 36cb5f8c9390410709054a710f97eda74781676c Mon Sep 17 00:00:00 2001 From: Yechan Kim <161688079+yechank-nvidia@users.noreply.github.com> Date: Thu, 5 Feb 2026 15:12:53 +0900 Subject: [PATCH] [https://nvbugs/5747920][fix] Fix multimodal serve test (#11296) Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com> --- examples/serve/aiperf_client_for_multimodal.sh | 2 +- tests/integration/test_lists/waives.txt | 1 - .../llmapi/apps/_test_trtllm_serve_multimodal_example.py | 7 +++++-- 3 files changed, 6 insertions(+), 4 deletions(-) diff --git a/examples/serve/aiperf_client_for_multimodal.sh b/examples/serve/aiperf_client_for_multimodal.sh index 14fbcfb614..cec9c0163d 100644 --- a/examples/serve/aiperf_client_for_multimodal.sh +++ b/examples/serve/aiperf_client_for_multimodal.sh @@ -2,7 +2,7 @@ aiperf profile \ -m Qwen2.5-VL-3B-Instruct \ - --tokenizer Qwen/Qwen2.5-VL-3B-Instruct \ + --tokenizer ${AIPERF_TOKENIZER_PATH:-Qwen/Qwen2.5-VL-3B-Instruct} \ --endpoint-type chat \ --random-seed 123 \ --image-width-mean 64 \ diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index 5ceb923db3..da2953bf7b 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -215,7 +215,6 @@ full:sm89/accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ full:sm89/accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp2] SKIP (https://nvbugs/5596337) unittest/executor/test_rpc.py::TestRpcCorrectness::test_incremental_task_async SKIP (https://nvbugs/5741476) test_e2e.py::test_trtllm_bench_llmapi_launch[pytorch_backend-llama-v3-llama3-8b] SKIP (https://nvbugs/5744432) -test_e2e.py::test_trtllm_serve_multimodal_example SKIP (https://nvbugs/5747920) cpp/test_multi_gpu.py::TestDisagg::test_symmetric_executor[gpt-2proc-mpi_kvcache-90] SKIP (https://nvbugs/5755941) examples/test_granite.py::test_llm_granite[granite-3.0-1b-a400m-instruct-bfloat16] SKIP (https://nvbugs/5608979) examples/test_granite.py::test_llm_granite[granite-3.0-2b-instruct-bfloat16] SKIP (https://nvbugs/5608979) diff --git a/tests/unittest/llmapi/apps/_test_trtllm_serve_multimodal_example.py b/tests/unittest/llmapi/apps/_test_trtllm_serve_multimodal_example.py index 490ccfb232..30c2b9d2c2 100644 --- a/tests/unittest/llmapi/apps/_test_trtllm_serve_multimodal_example.py +++ b/tests/unittest/llmapi/apps/_test_trtllm_serve_multimodal_example.py @@ -61,12 +61,15 @@ def example_root(): @pytest.mark.parametrize("exe, script", [("python3", "openai_chat_client_for_multimodal.py"), ("bash", "aiperf_client_for_multimodal.sh")]) -def test_trtllm_serve_examples(exe: str, script: str, +def test_trtllm_serve_examples(exe: str, script: str, model_name: str, server: RemoteOpenAIServer, example_root: str): client_script = os.path.join(example_root, script) + custom_env = os.environ.copy() + custom_env["AIPERF_TOKENIZER_PATH"] = get_model_path(model_name) # CalledProcessError will be raised if any errors occur subprocess.run([exe, client_script], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, - check=True) + check=True, + env=custom_env)