From 57ade4d0189087658a8b7a35fc19aed659bff6aa Mon Sep 17 00:00:00 2001 From: trtllm-agent <296075020+trtllm-agent@users.noreply.github.com> Date: Tue, 21 Jul 2026 04:09:52 -0700 Subject: [PATCH] [nvbugs/6478692][fix] Cap VideoMME EPD nano_omni_fp8 max_workers to 16 The nano_omni_fp8 EPDVariant did not override max_workers and defaulted to VideoMME.MAX_BATCH_SIZE=128, which oversubscribes the CPU-side multimodal preprocessing / hashing path on H100 and trips a native-library abort (SIGABRT) with many threads inside _media_to_raw_chw and apply_mm_hashes. The passing qwen3vl_2b variant already caps max_workers at 16 for the same reason. Cap the submission concurrency at 16 to match the working sibling variant, and remove the waiver so the test runs in CI again. Signed-off-by: trtllm-agent <296075020+trtllm-agent@users.noreply.github.com> --- tests/integration/defs/accuracy/test_epd_disagg_multimodal.py | 3 +++ tests/integration/test_lists/waives.txt | 1 - 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/tests/integration/defs/accuracy/test_epd_disagg_multimodal.py b/tests/integration/defs/accuracy/test_epd_disagg_multimodal.py index f32d0098b668..bfb3a0dba8d3 100644 --- a/tests/integration/defs/accuracy/test_epd_disagg_multimodal.py +++ b/tests/integration/defs/accuracy/test_epd_disagg_multimodal.py @@ -207,6 +207,9 @@ def nano_omni_fp8(cls) -> "EPDVariant": ), max_batch_size=64, expected_quant_algo=QuantAlgo.FP8, + # Default 128 workers trips a native-library abort in the CPU-side + # multimodal preprocessing path on H100 (nvbugs/6478692). + max_workers=16, ) @classmethod diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index 5417b0459701..1c7212ea9093 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -14,7 +14,6 @@ accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symm accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp2] SKIP (https://nvbugs/6428069) accuracy/test_disaggregated_serving.py::TestQwen3NextInstruct::test_auto_dtype[use_py_transceiver=False] SKIP (https://nvbugs/6535790) accuracy/test_disaggregated_serving.py::TestQwen3_30B_A3B::test_mixed_ctx_gen_model[ctxpp2gentp2] SKIP (https://nvbugs/5748664) -accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_fp8] SKIP (https://nvbugs/6478692) accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6367792) accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[nvfp4_ws8_80gb-trtllm] SKIP (https://nvbugs/6450341) accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_bf16_small[4] SKIP (https://nvbugs/6507114)