[TRTLLM-13613][test] Trim duplicated and dead multimodal accuracy tests from pre-merge CI (#15615)

Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>

Author: Wanli-Jiang

tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py

@@ -247,10 +247,9 @@ class TestNemotron_Nano_12B_V2_VL(LlmapiAccuracyTestHarness):
     @pytest.mark.parametrize(
         "enable_chunked_prefill,max_num_tokens",
         [
-            (False, MAX_NUM_TOKENS),
             (True, 1024),
         ],
-        ids=["full_budget", "forced_chunked_prefill"],
+        ids=["forced_chunked_prefill"],
     )
     def test_auto_dtype(self, enable_chunked_prefill, max_num_tokens):
         with LLM(
@@ -494,10 +493,9 @@ class TestQwen3VL(LlmapiAccuracyTestHarness):
     @pytest.mark.parametrize(
         "enable_chunked_prefill,max_num_tokens",
         [
-            (False, MAX_NUM_TOKENS),
             (True, 1024),
         ],
-        ids=["full_budget", "forced_chunked_prefill"],
+        ids=["forced_chunked_prefill"],
     )
     def test_auto_dtype(self, enable_chunked_prefill, max_num_tokens):
         with LLM(
@@ -543,8 +541,8 @@ class TestKimiK25(LlmapiAccuracyTestHarness):
     @pytest.mark.skip_less_device_memory(183000)
     @pytest.mark.parametrize(
         "ep_size,attention_dp",
-        [(1, False), (1, True), (8, False), (8, True)],
-        ids=["tp8", "tp8_attn_dp", "ep8", "dep8"],
+        [(8, True)],
+        ids=["dep8"],
     )
     def test_nvfp4(self, ep_size, attention_dp):
         """NVFP4 accuracy on MMMU benchmark (8x B200)."""
@@ -575,7 +573,6 @@ def test_nvfp4(self, ep_size, attention_dp):
 class TestMistralSmall24B(LlmapiAccuracyTestHarness):
     MODEL_NAME = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
     MODEL_PATH = f"{llm_models_root()}/Mistral-Small-3.1-24B-Instruct-2503"
-    MAX_NUM_TOKENS = 16384
 
     # NOTE: MMMU adds <|endoftext|> to the stop token.
     sampling_params = SamplingParams(
@@ -588,10 +585,9 @@ class TestMistralSmall24B(LlmapiAccuracyTestHarness):
     @pytest.mark.parametrize(
         "max_num_tokens",
         [
-            MAX_NUM_TOKENS,
             1024,
         ],
-        ids=["full_budget", "forced_chunked_prefill"],
+        ids=["forced_chunked_prefill"],
     )
     def test_auto_dtype(self, max_num_tokens):
         kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.75)
@@ -742,7 +738,7 @@ class TestNanoV3Omni(LlmapiAccuracyTestHarness):
                 ),
                 128,
                 QuantAlgo.MIXED_PRECISION,
-                (MMMU_TASK_SPEC, VOXPOPULI_TASK_SPEC, VIDEOMME_TASK_SPEC),
+                (MMMU_TASK_SPEC, VOXPOPULI_TASK_SPEC),
                 None,
                 marks=(skip_pre_blackwell,),
                 id="nvfp4",

tests/integration/test_lists/test-db/l0_b200.yml

@@ -54,7 +54,6 @@ l0_b200:
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a16_mxfp4[latency-TRTLLM]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention[target_sparsity_0.9-fp8kv=True]
   - accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=True]
-  - accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[nvfp4]
   - accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8_mmmu_encoder_cuda_graph]
   - accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[qwen3vl_2b_instruct]
   - accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_nvfp4]
@@ -355,6 +354,7 @@ l0_b200:
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-auto]
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-auto]
   - accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[bf16]
+  - accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[nvfp4]
   # ------------- VisualGen single-GPU tests ---------------
   - examples/visual_gen/test_visual_gen.py::test_visual_gen_quickstart
   - examples/visual_gen/test_visual_gen.py::test_visual_gen_api_walkthrough

tests/integration/test_lists/test-db/l0_dgx_b200.yml

@@ -164,7 +164,6 @@ l0_dgx_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_dsa_host_cache_offload[host_cache_offload_mtp3_no_adp] TIMEOUT (60)
   - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] TIMEOUT (60)
   - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8] TIMEOUT (60)
-  - accuracy/test_llm_api_pytorch_multimodal.py::TestKimiK25::test_nvfp4[dep8] TIMEOUT (120)
   - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp TIMEOUT (60)
   - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp_custom_op TIMEOUT (60)
   - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus[attention_dp_on-trtllm] TIMEOUT (60)
@@ -244,6 +243,7 @@ l0_dgx_b200:
   - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8_attn_dp] TIMEOUT (60)
   - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[ep8] TIMEOUT (60)
   - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[dep8] TIMEOUT (60)
+  - accuracy/test_llm_api_pytorch_multimodal.py::TestKimiK25::test_nvfp4[dep8] TIMEOUT (120)
   - accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_fp8[latency_moe_deepgemm] TIMEOUT (60)
   - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_parallelism[TP8_PP1] TIMEOUT (60)
   - test_e2e.py::test_deepseek_r1_mtp_bench TIMEOUT(60) # Cover https://nvbugs/5670108

tests/integration/test_lists/test-db/l0_l40s.yml

@@ -27,9 +27,6 @@ l0_l40s:
   - unittest/_torch/modeling/test_modeling_step3p7vl.py
   - test_e2e.py::test_ptp_scaffolding[DeepSeek-R1-Distill-Qwen-7B-DeepSeek-R1/DeepSeek-R1-Distill-Qwen-7B]
   - unittest/llmapi/apps/_test_openai_chat_multimodal.py::test_single_chat_session_image_embeds -m needs_l40s
-  # MMMU sanity check
-  - accuracy/test_llm_api_pytorch_multimodal.py::TestQwen2_5_VL_7B::test_auto_dtype
-  - accuracy/test_llm_api_pytorch_multimodal.py::TestVILA1_5_3B::test_auto_dtype
   # AutoDeploy: Nemotron-Nano-V3 on Ada uses flashinfer; trtllm has no
   # (E4M3 input, BF16 output, paged_kv, head_dim=128, sm_89) FMHA cubin.
   - accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-1-attn_dp_off-flashinfer]
@@ -85,8 +82,6 @@ l0_l40s:
   tests:
   - examples/test_gpt.py::test_llm_gpt2_next_prompt_tuning[use_cpp_session-tp1] # 10 mins
   - examples/test_gpt.py::test_llm_gpt2_next_prompt_tuning[use_py_session-tp1]
-  - examples/test_multimodal.py::test_llm_multimodal_general[video-neva-pp:1-tp:1-bfloat16-bs:1-cpp_e2e:False-nb:1]
-  - examples/test_multimodal.py::test_llm_multimodal_general[kosmos-2-pp:1-tp:1-float16-bs:1-cpp_e2e:True-nb:1]
   - examples/test_nemotron.py::test_llm_nemotron_3_8b_1gpu[bfloat16-fp8] # 18mins
   - examples/test_whisper.py::test_llm_whisper_general[large-v3-disable_gemm_plugin-disable_attention_plugin-disable_weight_only-float16-nb:1-use_python_runtime] # 8 mins
   - unittest/trt/attention/test_bert_attention.py # 12 mins