From a74d3bdc58d73ea7fc2410949a43a54876f53814 Mon Sep 17 00:00:00 2001 From: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com> Date: Thu, 7 May 2026 04:26:35 -0700 Subject: [PATCH 1/2] [nvbugs/6143787][fix] Lower KV cache fraction for TestQwen3VL to avoid OOM on L40S TestQwen3VL.test_auto_dtype had no KvCacheConfig override, so the default free_gpu_memory_fraction (~0.9) caused OOM during attention workspace allocation on 48GB GPUs. Match the sister TestQwen2_5_VL_7B pattern by setting free_gpu_memory_fraction=0.6. Remove the corresponding waiver. Signed-off-by: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com> --- .../defs/accuracy/test_llm_api_pytorch_multimodal.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py b/tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py index 7a84a910fc9a..4e248a59e1ff 100644 --- a/tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py +++ b/tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py @@ -480,6 +480,8 @@ class TestQwen3VL(LlmapiAccuracyTestHarness): max_tokens=MAX_NUM_TOKENS, truncate_prompt_tokens=MMMU.MAX_INPUT_LEN, stop="<|endoftext|>" ) + kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.6) + @pytest.mark.parametrize( "enable_chunked_prefill,max_num_tokens", [ @@ -493,6 +495,7 @@ def test_auto_dtype(self, enable_chunked_prefill, max_num_tokens): self.MODEL_PATH, enable_chunked_prefill=enable_chunked_prefill, max_num_tokens=max_num_tokens, + kv_cache_config=self.kv_cache_config, ) as llm: task = MMMU(self.MODEL_NAME) task.evaluate(llm, sampling_params=self.sampling_params) From 026ead07c2f79d50cbda24bc647ee63ab66cea1c Mon Sep 17 00:00:00 2001 From: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com> Date: Thu, 14 May 2026 23:29:16 -0700 Subject: [PATCH 2/2] [nvbugs/6143787][chore] Remove stale waiver after fix Signed-off-by: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com> --- tests/integration/test_lists/waives.txt | 1 - 1 file changed, 1 deletion(-) diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index ca3e31360dc0..2907bdc7ea81 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -147,7 +147,6 @@ accuracy/test_llm_api_pytorch_multimodal.py::TestExaone4_5_33B::test_auto_dtype[ accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized SKIP (https://nvbugs/6215689) accuracy/test_llm_api_pytorch_multimodal.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm] SKIP (https://nvbugs/6181383) accuracy/test_llm_api_pytorch_multimodal.py::TestNemotron_Nano_12B_V2_VL::test_auto_dtype[forced_chunked_prefill] SKIP (https://nvbugs/6248744) -accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3VL::test_auto_dtype[forced_chunked_prefill] SKIP (https://nvbugs/6143787) accuracy/test_llm_api_pytorch_ray.py::TestLlama3_1_8BInstruct::test_pp2_ray SKIP (https://nvbugs/6094070) cpp/test_e2e.py::test_benchmarks[bart-90] SKIP (https://nvbugs/5550689) cpp/test_e2e.py::test_benchmarks[gpt-80] SKIP (https://nvbugs/5550689)