[None][test] Add DeepSeek-V4.1-Flash, GLM-5.3-Flash, Qwen3.8-Flash-Next, Kimi-K3 and MiniMax-M3 perf-sanity cases

jenkins/L0_Test.groovy

@@ -6677,10 +6677,12 @@ def launchTestJobs(pipeline, testFilter, globalVars)
         "GB300-4_GPUs-PyTorch-PerfSanity-1": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 1, 1, 4, 1, true, false],
         // PerfSanity post-merge tests
         "GB200-4_GPUs-PyTorch-PerfSanity-Post-Merge-1": ["auto:gb200-x4", "l0_gb200_multi_gpus_perf_sanity", 1, 1, 4],
-        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-1": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 1, 4, 4, 1, true, false],
-        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-2": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 2, 4, 4, 1, true, false],
-        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-3": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 3, 4, 4, 1, true, false],
-        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-4": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 4, 4, 4, 1, true, false],
+        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-1": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 1, 6, 4, 1, true, false],
+        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-2": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 2, 6, 4, 1, true, false],
+        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-3": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 3, 6, 4, 1, true, false],
+        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-4": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 4, 6, 4, 1, true, false],
+        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-5": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 5, 6, 4, 1, true, false],
+        "GB300-4_GPUs-PyTorch-PerfSanity-Post-Merge-6": ["auto:gb300-x4", "l0_gb300_multi_gpus_perf_sanity", 6, 6, 4, 1, true, false],
     ]
     SBSASlurmTestConfigs = cbtsResizeSplits(SBSASlurmTestConfigs)
     fullSet += SBSASlurmTestConfigs.keySet()
@@ -6798,7 +6800,7 @@ def launchTestJobs(pipeline, testFilter, globalVars)
         "GB300-8_GPUs-2_Nodes-PyTorch-PerfSanity-GEN1-NODE2-GPU8-Post-Merge",
         "auto:gb300-flex",
         "l0_gb300_multi_nodes_perf_sanity_gen1_node2_gpu8",
-        4,
+        6,
         8,
         2
     )
@@ -6807,7 +6809,7 @@ def launchTestJobs(pipeline, testFilter, globalVars)
         "GB300-16_GPUs-4_Nodes-PyTorch-PerfSanity-GEN1-NODE4-GPU16-Post-Merge",
         "auto:gb300-flex",
         "l0_gb300_multi_nodes_perf_sanity_gen1_node4_gpu16",
-        1,
+        2,
         16,
         4
     )
@@ -6820,12 +6822,12 @@ def launchTestJobs(pipeline, testFilter, globalVars)
         32,
         8
     )
-    // Nemotron-Ultra-V3 8k64k con1: ctx1 (1 node, 4 GPUs) + gen1 tep4 (1 node, 4 GPUs) = 8 GPUs
+    // Nemotron-Ultra-V3 8k64k con1 and GLM-5.3-Flash FP8 8k1k con16: ctx1 (1 node, 4 GPUs) + gen1 tep4 (1 node, 4 GPUs) = 8 GPUs
     multiNodesSBSAConfigs += buildStageConfigs(
         "GB300-8_GPUs-2_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE1-GPU4-Post-Merge",
         "auto:gb300-flex",
         "l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4",
-        1,
+        2,
         8,
         2
     )

tests/integration/defs/.test_durations

@@ -900,18 +900,28 @@
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 2753.409214285714,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 416.70321428571424,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 1575.1635714285715,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm53-flash-fp8_8k1k_con16_ctx1_tep4_gen1_tep4_eplb0_mtp3_ccb-NIXL]": 419.311,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_minimax-m3-fp4_150k1k_con1_ctx1_tp2_gen1_tp4_eplb0_eagle3_ccb-NIXL]": 1059.366,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_nemotron-ultra-v3-fp4_50k2k_con1197_ctx16_dep4_gen1_dep8_eplb0_mtp3_ccb-NIXL]": 1168.722642857143,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_nemotron-ultra-v3-fp4_50k2k_con12_ctx1_dep4_gen6_tep4_eplb0_mtp6_ccb-NIXL]": 577.7239285714285,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_nemotron-ultra-v3-fp4_8k64k_con1_ctx1_dep4_gen1_tep4_eplb0_mtp5_ccb-NIXL]": 408.025,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_nemotron-ultra-v3-fp4_8k64k_con9832_ctx1_dep4_gen8_dep8_eplb0_mtp3_ccb-NIXL]": 2200.323,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v41_flash_grace_blackwell-v41_flash_dep4_8k1k]": 624.423,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v41_flash_grace_blackwell-v41_flash_tep4_8k1k]": 635.625,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v41_flash_grace_blackwell-v41_flash_tep4_dspark_8k1k]": 701.992,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_235b_a22b_fp8_hopper-qwen3_235b_a22b_fp8_tp4_ep4_cutlass_8k1k]": 944.3419230769231,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_32b_fp8_hopper-qwen3_32b_fp8_tp2_6k1k]": 703.392,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_deepseek_v41_flash_nvfp4_grace_blackwell-v41_flash_nvfp4_dep4_dspark5_8k1k]": 904.2026666666666,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_deepseek_v41_flash_nvfp4_grace_blackwell-v41_flash_nvfp4_tep4_dspark5_8k1k]": 767.533,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_fp8_grace_blackwell-glm53_flash_fp8_dep4_mtp3_fp8kv_8k1k]": 606.433,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_fp8_grace_blackwell-glm53_flash_fp8_tp4_mtp3_8k1k]": 379.83566666666667,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_fp8_grace_blackwell-glm53_flash_fp8_tp4_mtp3_isl13312_osl33]": 483.808,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_nvfp4_grace_blackwell-glm53_flash_nvfp4_tp4_mtp3_8k1k]": 541.238,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_nvfp4_grace_blackwell-glm53_flash_nvfp4_tp4_mtp3_isl13312_osl33]": 967.477,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_nemotron_ultra_v3_fp4_grace_blackwell-nemotron_ultra_v3_fp4_dep4_mtp0_50k2k]": 1863.1629285714287,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_nemotron_ultra_v3_fp4_grace_blackwell-nemotron_ultra_v3_fp4_dep4_mtp3_50k2k]": 780.4455,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_nemotron_ultra_v3_fp4_grace_blackwell-nemotron_ultra_v3_fp4_tep4_mtp6_50k2k]": 1339.9197142857142,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_qwen3_8_flash_next_fp4_grace_blackwell-qwen3_8_flash_next_fp4_mv2_tp1_mtp3_8k1k]": 1011.102,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gemma4_26b_a4b_nvfp4_blackwell-gemma4_26b_a4b_nvfp4_tp1_1k1k]": 289.63309090909087,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 284.1654117647059,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 228.4835294117647,
@@ -923,6 +933,12 @@
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 794.9620666666666,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 1609.1394285714287,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 411.37428571428575,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm53-flash-fp8_8k1k_con16_ctx1_tep4_gen1_tep4_eplb0_mtp3_ccb-NIXL]": 433.262,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_kimi-k3-fp4-dspark_160k1k_con16_ctx16_dwdp16_gen1_helix8_bs16_dspark5_ccb-NIXL]": 741.683,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_kimi-k3-fp4-dspark_160k1k_con32_ctx16_dwdp16_gen1_helix16_bs32_dspark6_ccb-NIXL]": 886.427,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_minimax-m3-fp4_150k1k_con128_ctx1_tp2_gen1_dep8_eplb0_eagle3_ccb-NIXL]": 913.965,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_minimax-m3-fp4_150k1k_con1_ctx1_tp2_gen1_tp4_eplb0_eagle3_ccb-NIXL]": 1187.22,
+   "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_minimax-m3-fp4_150k1k_con64_ctx1_tp2_gen1_tp4_eplb0_eagle3_ccb-NIXL]": 1759.201,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_nemotron-ultra-v3-fp4_50k2k_con12_ctx1_dep4_gen6_tep4_eplb0_mtp6_ccb-NIXL]": 1415.384142857143,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_nemotron-ultra-v3-fp4_8k64k_con1_ctx1_dep4_gen1_tep4_eplb0_mtp5_ccb-NIXL]": 1434.301857142857,
    "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 655.066,
@@ -941,6 +957,7 @@
    "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1202.3937142857144,
    "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1021.8967857142857,
    "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 3149.737866666667,
+   "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm53-flash-fp8_8k1k_con16_ctx1_tep4_gen1_tep4_eplb0_mtp3_ccb-NIXL]": 504.041,
    "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_nemotron-ultra-v3-fp4_50k2k_con12_ctx1_dep4_gen6_tep4_eplb0_mtp6_ccb-NIXL]": 1649.4002142857144,
    "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 519.6357058823529,
    "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 342.2766470588235,

tests/integration/defs/perf/_model_paths.py

@@ -92,6 +92,7 @@
     # GLM-5.2 NVFP4 (MoE, MLA + DSA on the DeepSeek-V3.2 code path)
     "glm_5.2_fp4": "GLM-5.2-NVFP4",
     "glm_5.3_flash_fp8": "GLM-5.3-Flash",
+    "glm_5.3_flash_nvfp4": "GLM-5.3-Flash-NVFP4",
     # MiniMax-M3 NVFP4: MXFP8 base layers with NVFP4 routed experts
     # (the MXFP8 checkpoint is registered as "minimax_m3_mxfp8" above).
     "minimax_m3_fp4": "MiniMax-M3-NVFP4",

tests/integration/test_lists/test-db/l0_gb300_multi_gpus_perf_sanity.yml

@@ -41,11 +41,27 @@ l0_gb300_multi_gpus_perf_sanity:
   # deepseek-v4-pro-fp4 8k1k
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL] TIMEOUT (45)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL] TIMEOUT (125)
+  # glm-5.3-flash-fp8 8k1k con16
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm53-flash-fp8_8k1k_con16_ctx1_tep4_gen1_tep4_eplb0_mtp3_ccb-NIXL] TIMEOUT (30)
+  # minimax-m3-fp4 150k prompts, TP2 ctx worker
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_minimax-m3-fp4_150k1k_con1_ctx1_tp2_gen1_tp4_eplb0_eagle3_ccb-NIXL] TIMEOUT (40)
   # deepseek-v4.1-flash-fp4 8k1k dspark (aggregated)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_deepseek_v41_flash_nvfp4_grace_blackwell-v41_flash_nvfp4_tep4_dspark5_8k1k] TIMEOUT (30)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_deepseek_v41_flash_nvfp4_grace_blackwell-v41_flash_nvfp4_dep4_dspark5_8k1k] TIMEOUT (35)
+  # deepseek-v4.1-flash 8k1k (aggregated)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v41_flash_grace_blackwell-v41_flash_tep4_8k1k] TIMEOUT (30)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v41_flash_grace_blackwell-v41_flash_dep4_8k1k] TIMEOUT (30)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v41_flash_grace_blackwell-v41_flash_tep4_dspark_8k1k] TIMEOUT (30)
   # glm-5.3-flash-fp8 tp4 mtp3 8k1k (aggregated)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_fp8_grace_blackwell-glm53_flash_fp8_tp4_mtp3_8k1k] TIMEOUT (30)
+  # glm-5.3-flash-fp8 attention dp4 mtp3 fp8 kv 8k1k and long prompt (aggregated)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_fp8_grace_blackwell-glm53_flash_fp8_dep4_mtp3_fp8kv_8k1k] TIMEOUT (30)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_fp8_grace_blackwell-glm53_flash_fp8_tp4_mtp3_isl13312_osl33] TIMEOUT (30)
+  # glm-5.3-flash-nvfp4 tp4 mtp3 8k1k (aggregated)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_nvfp4_grace_blackwell-glm53_flash_nvfp4_tp4_mtp3_8k1k] TIMEOUT (30)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_glm53_flash_nvfp4_grace_blackwell-glm53_flash_nvfp4_tp4_mtp3_isl13312_osl33] TIMEOUT (35)
+  # qwen3.8-flash-next-nvfp4 modeling_v2 tp1 mtp3 8k1k (aggregated, 1 GPU)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gb300_qwen3_8_flash_next_fp4_grace_blackwell-qwen3_8_flash_next_fp4_mv2_tp1_mtp3_8k1k] TIMEOUT (35)
   # nemotron-ultra-v3-fp4 8k64k (ctx_only)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_nemotron-ultra-v3-fp4_8k64k_con1_ctx1_dep4_gen1_tep4_eplb0_mtp5_ccb-NIXL] TIMEOUT (30)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_nemotron-ultra-v3-fp4_8k64k_con9832_ctx1_dep4_gen8_dep8_eplb0_mtp3_ccb-NIXL] TIMEOUT (60)
@@ -60,3 +76,8 @@ l0_gb300_multi_gpus_perf_sanity:
   # nemotron-ultra-v3-fp4
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_nemotron-ultra-v3-fp4_8k64k_con1_ctx1_dep4_gen1_tep4_eplb0_mtp5_ccb-NIXL] TIMEOUT (40)
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_nemotron-ultra-v3-fp4_50k2k_con12_ctx1_dep4_gen6_tep4_eplb0_mtp6_ccb-NIXL] TIMEOUT (40)
+  # glm-5.3-flash-fp8 8k1k con16
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm53-flash-fp8_8k1k_con16_ctx1_tep4_gen1_tep4_eplb0_mtp3_ccb-NIXL] TIMEOUT (30)
+  # minimax-m3-fp4 eagle3 150k1k con1 and con64, TP4 gen worker
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_minimax-m3-fp4_150k1k_con1_ctx1_tp2_gen1_tp4_eplb0_eagle3_ccb-NIXL] TIMEOUT (35)
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_minimax-m3-fp4_150k1k_con64_ctx1_tp2_gen1_tp4_eplb0_eagle3_ccb-NIXL] TIMEOUT (45)

tests/integration/test_lists/test-db/l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4.yml

@@ -16,3 +16,5 @@ l0_gb300_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node1_gpu4:
   tests:
   # nemotron-ultra-v3-fp4 8k64k con1
   - perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_nemotron-ultra-v3-fp4_8k64k_con1_ctx1_dep4_gen1_tep4_eplb0_mtp5_ccb-NIXL] TIMEOUT (45)
+  # glm-5.3-flash-fp8 8k1k con16
+  - perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm53-flash-fp8_8k1k_con16_ctx1_tep4_gen1_tep4_eplb0_mtp3_ccb-NIXL] TIMEOUT (30)

tests/integration/test_lists/test-db/l0_gb300_multi_nodes_perf_sanity_gen1_node2_gpu8.yml

@@ -22,3 +22,7 @@ l0_gb300_multi_nodes_perf_sanity_gen1_node2_gpu8:
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL] TIMEOUT (45)
   # glm-5-fp4 8k1k con1024
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL] TIMEOUT (30)
+  # kimi-k3-fp4 dspark 160k1k con16, helix CP8 gen worker
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_kimi-k3-fp4-dspark_160k1k_con16_ctx16_dwdp16_gen1_helix8_bs16_dspark5_ccb-NIXL] TIMEOUT (30)
+  # minimax-m3-fp4 eagle3 150k1k con128, DEP8 gen worker
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_minimax-m3-fp4_150k1k_con128_ctx1_tp2_gen1_dep8_eplb0_eagle3_ccb-NIXL] TIMEOUT (35)

tests/integration/test_lists/test-db/l0_gb300_multi_nodes_perf_sanity_gen1_node4_gpu16.yml

@@ -16,3 +16,5 @@ l0_gb300_multi_nodes_perf_sanity_gen1_node4_gpu16:
   tests:
   # deepseek-v4-pro-fp4 8k1k con666
   - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL] TIMEOUT (30)
+  # kimi-k3-fp4 dspark 160k1k con32, helix CP16 gen worker
+  - perf/test_perf_sanity.py::test_e2e[aggr_upload-gen_only_no_context-gb300_kimi-k3-fp4-dspark_160k1k_con32_ctx16_dwdp16_gen1_helix16_bs32_dspark6_ccb-NIXL] TIMEOUT (30)

tests/scripts/perf-sanity/aggregated/deepseek_v41_flash_grace_blackwell.yaml

@@ -10,7 +10,9 @@
 # derivation changes (see disagg_config_ctxtp4_gentp4_deepseek_v41.yaml):
 #   sparse_attention_config   auto-derived from architectures[0].
 #   kv_cache_config.dtype     the CSA2 TRTLLM backend stages BF16 KV and rejects
-#                             anything but auto/bfloat16.
+#                             anything but auto/bfloat16. Written out as `auto`
+#                             (the default) only because the perf harness
+#                             records a missing dtype as fp8.
 #   tokens_per_block          model default (128 == sliding_window).
 #   moe_config.backend        V4.1's experts are MXFP4; the model selects its
 #                             own MoE path.
@@ -49,6 +51,7 @@ server_configs:
       enable_padding: true
       batch_sizes: [1, 2, 4, 8]
     kv_cache_config:
+      dtype: auto
       enable_block_reuse: false
       free_gpu_memory_fraction: 0.8
     client_configs:
@@ -81,6 +84,7 @@ server_configs:
       enable_padding: true
       batch_sizes: [1, 2, 4, 8, 16, 32]
     kv_cache_config:
+      dtype: auto
       enable_block_reuse: false
       free_gpu_memory_fraction: 0.8
     client_configs:
@@ -110,6 +114,7 @@ server_configs:
       enable_padding: true
       batch_sizes: [1, 2, 4, 8]
     kv_cache_config:
+      dtype: auto
       enable_block_reuse: false
       free_gpu_memory_fraction: 0.8
     speculative_config: