[TRTLLM-15040][test] Prune legacy Llama and Nemotron tests

tests/integration/defs/.test_durations

@@ -74,7 +74,6 @@
    "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[deepseek-ai_DeepSeek-R1-0528-True]": 807.0042,
    "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[google_gemma-3-1b-it-False]": 54.751599999999996,
    "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.1-8B-Instruct-False]": 43.825,
-   "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.3-70B-Instruct-False]": 141.4636,
    "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Codestral-22B-v0.1-False]": 85.2512,
    "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Ministral-8B-Instruct-2410-False]": 58.8104,
    "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_DeepSeek-R1-0528-NVFP4-v2-True]": 1474.047,
@@ -461,11 +460,6 @@
    "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=False]": 774.0882071823204,
    "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=True]": 699.8562759562841,
    "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B_Instruct_RocketKV::test_auto_dtype": 922.8317,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp4_tp2pp2[torch_compile=True-enable_gemm_allreduce_fusion=False]": 535.9970232558139,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=False]": 607.9050751879699,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=True]": 807.2856390977444,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=False]": 633.9083157894737,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=True]": 815.445175572519,
    "accuracy/test_llm_api_pytorch.py::TestMiniMaxM2::test_4gpus[attention_dp=False-cuda_graph=True-overlap_scheduler=True-tp_size=4-ep_size=4]": 292.3591666666667,
    "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=True]": 863.916775510204,
    "accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm]": 476.5565,
@@ -917,7 +911,6 @@
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 509.8764444444444,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp2_1k8k]": 459.85290000000003,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_deepseek_v3_lite-v3lite_fp8_bs8_128_256]": 596.7645321782178,
-   "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b-llama8b_fp16_bs8_128_256]": 257.1823105134474,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_8k1k]": 698.3583333333333,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_mtp3_8k1k]": 531.4513333333334,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tep4_mtp3_8k1k]": 404.2888333333333,
@@ -968,8 +961,6 @@
    "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan22_i2v_a14b_blackwell-wan22_i2v_a14b_nvfp4_trtllm_cfg2_ulysses4]": 342.01559999999995,
    "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp1_4instances]": 161.67143564356437,
    "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp2_2instances]": 103.54573267326732,
-   "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-GUARANTEED_NO_EVICT-pytorch-stress-test]": 710.0681999999999,
-   "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-MAX_UTILIZATION-pytorch-stress-test]": 702.3580999999999,
    "test_doc.py::test_url_validity": 49.48444444444444,
    "test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 54.35407142857143,
    "test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3_depth_1_tree[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 55.181,
@@ -1080,7 +1071,6 @@
    "unittest/_torch/modeling -k \"modeling_llama\"": 116.07080075662043,
    "unittest/_torch/modeling -k \"modeling_mixtral\"": 72.60693384223919,
    "unittest/_torch/modeling -k \"modeling_nemotron_nano_v2_vl\"": 421.9780921658986,
-   "unittest/_torch/modeling -k \"modeling_nemotron_nas\"": 42.0892577092511,
    "unittest/_torch/modeling -k \"modeling_out_of_tree\"": 144.867929456112,
    "unittest/_torch/modeling -k \"modeling_phi3\"": 35.700971046770604,
    "unittest/_torch/modeling -k \"modeling_siglip\"": 204.05294545454547,
@@ -1494,7 +1484,6 @@
    "unittest/llmapi/test_llm_pytorch.py -m \"part1\"": 249.2021185086551,
    "unittest/llmapi/test_llm_pytorch.py -m \"part2\"": 349.5203164893617,
    "unittest/llmapi/test_llm_pytorch.py -m \"part3\"": 279.06796533333335,
-   "unittest/llmapi/test_llm_pytorch.py::test_nemotron_nas_lora": 194.5492,
    "unittest/llmapi/test_llm_quant.py": 21.754477644492912,
    "unittest/llmapi/test_llm_telemetry.py": 56.14217090620032,
    "unittest/llmapi/test_llm_telemetry.py::TestTelemetryArchitectureExtraction": 61.80126423690205,

tests/integration/defs/.test_durations_aws_dfw

@@ -4,7 +4,6 @@
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=0]": 293.95560641004704,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 275.8909912491217,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 266.709816042101,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp4_tp2pp2[torch_compile=True-enable_gemm_allreduce_fusion=False]": 908.2052957660053,
    "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 641.9124943269417,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=0]": 287.03202630905434,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=2]": 350.3160223159939,
@@ -15,8 +14,6 @@
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 430.816315329168,
    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-cutlass-auto]": 1067.7517247761134,
    "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus_online_eplb[fp8]": 813.0690455089789,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=False]": 702.5942377618048,
-   "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=False]": 848.7312017090153,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_online_eplb[mtp_nextn=2-moe_backend=CUTLASS]": 413.68237058399245,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 319.0723833630327,
    "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 332.56335388694424,

tests/integration/defs/accuracy/references/cnn_dailymail.yaml

@@ -34,12 +34,6 @@ gpt2-medium:
     accuracy: 22.249
 gpt-next:
   - accuracy: 25.516
-microsoft/phi-2:
-  - accuracy: 31.255
-bigcode/starcoder2-7b:
-  - accuracy: 26.611
-  - quant_algo: FP8
-    accuracy: 26.611
 state-spaces/mamba-130m-hf:
   - accuracy: 19.470
 lmsys/vicuna-7b-v1.3:
@@ -71,13 +65,6 @@ TinyLlama/TinyLlama-1.1B-Chat-v1.0:
     accuracy: 27.882
   - extra_acc_spec: pp_size=4
     accuracy: 15.123
-meta-llama/Meta-Llama-3-8B-Instruct:
-  - accuracy: 34.957
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 34.737
-  - quant_algo: W8A16_GPTQ
-    accuracy: 34.858
 meta-llama/Llama-3.1-8B:
   - accuracy: 24.360
   - quant_algo: W8A8_SQ_PER_CHANNEL_PER_TOKEN_PLUGIN
@@ -122,48 +109,6 @@ meta-llama/Llama-3.1-8B-Instruct:
   - quant_algo: FP8
     extra_acc_spec: beam_width=2
     accuracy: 31.201
-meta-llama/Llama-3.2-1B:
-  - accuracy: 27.427
-  - quant_algo: W8A8_SQ_PER_CHANNEL_PER_TOKEN_PLUGIN
-    accuracy: 27.931
-  - quant_algo: W8A8_SQ_PER_CHANNEL
-    accuracy: 25.631
-  - quant_algo: W4A16_AWQ
-    accuracy: 25.028
-  - quant_algo: W4A16_AWQ
-    kv_cache_quant_algo: INT8
-    accuracy: 24.354
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 27.029
-  - quant_algo: FP8
-    accuracy: 27.029
-  - quant_algo: FP8_PER_CHANNEL_PER_TOKEN
-    accuracy: 27.257
-  - quant_algo: FP8_PER_CHANNEL_PER_TOKEN
-    extra_acc_spec: meta_recipe
-    accuracy: 27.614
-  - extra_acc_spec: max_attention_window_size=960
-    accuracy: 27.259
-  - extra_acc_spec: max_attention_window_size=960;beam_width=4
-    accuracy: 0
-meta-llama/Llama-3.2-3B:
-  - accuracy: 25.495
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 33.629
-meta-llama/Llama-3.3-70B-Instruct:
-  - quant_algo: FP8
-    spec_dec_algo: Eagle
-    accuracy: 33.244
-  - quant_algo: FP8
-    spec_dec_algo: Eagle3
-    accuracy: 33.244
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 34.383
-  - quant_algo: FP8
-    accuracy: 34.927
 mistralai/Mistral-Small-3.1-24B-Instruct-2503:
   - accuracy: 29.20
   - quant_algo: FP8
@@ -239,5 +184,3 @@ Qwen3/Qwen3-8B:
   - quant_algo: FP8_BLOCK_SCALES
     accuracy: 30
   - accuracy: 30
-nvidia/Llama-3_3-Nemotron-Super-49B-v1:
-  - accuracy: 34.003

tests/integration/defs/accuracy/references/gpqa_diamond.yaml

@@ -1,16 +1,3 @@
-meta-llama/Llama-3.3-70B-Instruct:
-  - accuracy: 45.96
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 45.55
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 48.03
-  - quant_algo: FP8
-    accuracy: 48.03
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 48.03
 deepseek-ai/DeepSeek-R1:
   - quant_algo: NVFP4
     accuracy: 70.45
@@ -35,30 +22,6 @@ deepseek-ai/DeepSeek-V3.2-Exp:
   - quant_algo: NVFP4
     spec_dec_algo: MTP
     accuracy: 80.0
-nvidia/Llama-3_3-Nemotron-Super-49B-v1:
-  - accuracy: 44.95
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 42.42
-    # GPQA diamond only contains 198 samples, so the score tends to have large variance.
-    # We repeated evaluation 7 times to choose a lower bound score for FP8, 42.42.
-    # random_seed=0: 47.98
-    # random_seed=1: 42.42
-    # random_seed=2: 52.02
-    # random_seed=3: 51.52
-    # random_seed=4: 48.48
-    # random_seed=5: 47.47
-    # random_seed=6: 45.96
-nvidia/Llama-3.1-Nemotron-Nano-8B-v1:
-  - accuracy: 40.40
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 39.39
-nvidia/Llama-3_1-Nemotron-Ultra-253B-v1:
-  - accuracy: 58.08
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 57.07
 GPT-OSS/120B-MXFP4:
   - accuracy: 65.0
   - spec_dec_algo: Eagle

tests/integration/defs/accuracy/references/gsm8k.yaml

@@ -33,24 +33,6 @@ meta-llama/Llama-3.1-8B-Instruct:
   - quant_algo: NVFP4
     kv_cache_quant_algo: FP8
     accuracy: 66.03
-meta-llama/Llama-3.3-70B-Instruct:
-  - accuracy: 83.78
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 87.33
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 90.30
-  - quant_algo: FP8
-    accuracy: 90.30
-meta-llama/Llama-4-Scout-17B-16E-Instruct:
-  - accuracy: 89.70
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 88.61
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 89.45
 deepseek-ai/DeepSeek-V3-Lite:
   - accuracy: 64.74
   - quant_algo: NVFP4
@@ -318,11 +300,6 @@ moonshotai/Kimi-K3:
   # SA spec dec is lossless; scores match the baseline within noise.
   - spec_dec_algo: SA
     accuracy: 96.5
-nvidia/Llama-3_3-Nemotron-Super-49B-v1:
-  - accuracy: 92.57
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 92.42
 nvidia/Nemotron-MOE:
   - accuracy: 88.249
   - quant_algo: FP8
@@ -331,16 +308,6 @@ nvidia/Nemotron-MOE:
   - quant_algo: NVFP4
     kv_cache_quant_algo: FP8
     accuracy: 63.268
-nvidia/Llama-3.1-Nemotron-Nano-8B-v1:
-  - accuracy: 37.15
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 28.39
-nvidia/Llama-3_1-Nemotron-Ultra-253B-v1:
-  - accuracy: 94.43
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 94.16
 google/gemma-3-1b-it:
   - accuracy: 25.52 # score getting from lm-eval with HF implementation
   - quant_algo: FP8
@@ -433,8 +400,6 @@ GPT-OSS/20B-NVFP4:
     accuracy: 85.0
 ByteDance-Seed/Seed-OSS-36B-Instruct:
   - accuracy: 90.8
-bigcode/starcoder2-7b:
-  - accuracy: 26.5
 bigcode/starcoder2-15b:
   - accuracy: 54.5
 poolside/laguna-XS.2:

tests/integration/defs/accuracy/references/mmlu.yaml

@@ -1,8 +1,3 @@
-meta-llama/Meta-Llama-3-8B-Instruct:
-  - accuracy: 67.74
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 63.47
 meta-llama/Llama-3.1-8B:
   - accuracy: 66.06
   - quant_algo: NVFP4
@@ -35,59 +30,6 @@ meta-llama/Llama-3.1-8B-Instruct:
   - quant_algo: NVFP4
     kv_cache_quant_algo: FP8
     accuracy: 65.11
-meta-llama/Llama-3.2-1B:
-  - quant_algo: W8A8_SQ_PER_CHANNEL_PER_TOKEN_PLUGIN
-    accuracy: 32.72
-  - quant_algo: W8A8_SQ_PER_CHANNEL
-    accuracy: 32.07
-  - quant_algo: W4A16_AWQ
-    accuracy: 30.56
-  - quant_algo: W4A16_AWQ
-    kv_cache_quant_algo: INT8
-    accuracy: 31.29
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 31.02
-  - quant_algo: FP8_PER_CHANNEL_PER_TOKEN
-    accuracy: 33.97
-  - quant_algo: FP8_PER_CHANNEL_PER_TOKEN
-    extra_acc_spec: meta_recipe
-    accuracy: 33.87
-  - extra_acc_spec: max_attention_window_size=960
-    accuracy: 32.82
-meta-llama/Llama-3.2-3B:
-  - accuracy: 57.92
-  - spec_dec_algo: Eagle3
-    accuracy: 57.92
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 60.60
-meta-llama/Llama-3.3-70B-Instruct:
-  - accuracy: 81.31
-  - spec_dec_algo: Eagle3
-    accuracy: 81.31
-  - quant_algo: FP8
-    spec_dec_algo: Eagle
-    accuracy: 81.31
-  - quant_algo: FP8
-    spec_dec_algo: Eagle3
-    accuracy: 81.31
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 78.78
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 80.40
-  - quant_algo: FP8
-    accuracy: 80.40
-meta-llama/Llama-4-Scout-17B-16E-Instruct:
-  - accuracy: 80.00
-  - quant_algo: NVFP4
-    kv_cache_quant_algo: FP8
-    accuracy: 79.60
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 78.58
 mistralai/Mistral-Small-3.1-24B-Instruct-2503:
   - accuracy: 81.7
   - quant_algo: FP8
@@ -256,26 +198,7 @@ moonshotai/Kimi-K2.5:
   - quant_algo: NVFP4
     kv_cache_quant_algo: FP8
     accuracy: 89.67
-nvidia/Llama-3_3-Nemotron-Super-49B-v1:
-  - accuracy: 79.43
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 79.26
-nvidia/Llama-3.1-Nemotron-Nano-8B-v1:
-  - accuracy: 57.97
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 57.12
-bigcode/starcoder2-7b:
-  - accuracy: 41.35
-  - quant_algo: FP8
-    accuracy: 41.35
 # TODO: update once https://nvbugs/5393849 is fixed.
-nvidia/Llama-3_1-Nemotron-Ultra-253B-v1:
-  - accuracy: 83.70
-  - quant_algo: FP8
-    kv_cache_quant_algo: FP8
-    accuracy: 83.36
 mistralai/Ministral-8B-Instruct-2410:
   - accuracy: 66.35
   - quant_algo: FP8