{
  "title": "GLM-5.3-Flash SGLang rescue on four H200 NVL GPUs",
  "completed_utc": "2026-10-12",
  "published": "2026-10-13",
  "audit_errors": [],
  "caveats": [
    "Finite short-chat tests, not a long-duration soak or long-context qualification.",
    "Cache-enabled repeats reuse prompts; report each repeat as well as medians.",
    "Two isolated GPU quads ran concurrently on the same host; not an isolated performance A/B.",
    "Output-token counts include reasoning tokens; 256-token generation limit."
  ],
  "sources": [
    {
      "url": "https://github.com/sgl-project/sglang/blob/main/docs/src/snippets/configs/zai-org/glm-5.3-flash.jsx",
      "use": "Official H200 TP8/EP8 and EAGLE5/topk1/6-token guidance; adapted to a four-GPU NVLink group, not an exact official recipe reproduction."
    },
    {
      "url": "https://github.com/sgl-project/sglang/blob/main/docs/cookbook/autoregressive/GLM/GLM-5.3-Flash.mdx",
      "use": "Keep prefix caching enabled; distinguish KV and KDA state pools."
    },
    {
      "url": "https://github.com/sgl-project/sglang/issues/37712",
      "use": "H200 long-prefill OOM report motivated 4096 prefill chunks and 0.78 static memory; original failure here is illegal memory access, not established OOM."
    },
    {
      "url": "https://docs.nvidia.com/dynamo/v1.5.0/recipes/glm-5-3-flash",
      "use": "Additional TP8/MTP7/FLASHMLA_SPARSE direction; not tested in this round."
    }
  ],
  "archive": {
    "name": "h200-glm-sglang-rescue-20261012.tar.zst",
    "sha256": "7fa12f22c16e2ac71f8bc4dc9d749d4cefafa49b6b2d8214a43518e2e716911e"
  },
  "runs": [
    {
      "name": "reproduce",
      "state": "failed",
      "raw_completed": 2,
      "raw_failed": 1,
      "cuda_illegal_memory": true,
      "quality": {
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {},
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 1,
        "gpus": 4,
        "cuda_devices": "0,1,2,3",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 8192,
        "memory_fraction": 0.85,
        "prefix_caching": false,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 64,
        "engine_kv_tokens": 3343552,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": []
    },
    {
      "name": "no-cuda-graphs",
      "state": "diagnostic_pass",
      "raw_completed": 808,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {
        "structured_after-first-sweep.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 1,
          "output_tok_s": 7.127764236328813,
          "ttft_p99_s": 22.738563086616338,
          "latency_p99_s": 54.14256675332115
        },
        "16": {
          "repeats": 1,
          "output_tok_s": 83.34489949242561,
          "ttft_p99_s": 24.560182506071868,
          "latency_p99_s": 55.63449692468246
        },
        "64": {
          "repeats": 1,
          "output_tok_s": 351.8727573605711,
          "ttft_p99_s": 15.630398686900977,
          "latency_p99_s": 55.075061435559476
        },
        "128": {
          "repeats": 1,
          "output_tok_s": 381.58544791469023,
          "ttft_p99_s": 66.34375637894759,
          "latency_p99_s": 100.22422587619978
        }
      },
      "reason": "One complete C1/C16/C64/C128 sweep plus before/after32 correctness checks passed. Further repeats intentionally omitted because graph-disabled single-request throughput was about7tok/s; diagnostic fallback, not a performance candidate.",
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 1,
        "gpus": 4,
        "cuda_devices": "0,1,2,3",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 8192,
        "memory_fraction": 0.85,
        "prefix_caching": false,
        "disable_cuda_graph": true,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 64,
        "engine_kv_tokens": 3343552,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": [
        {
          "file": "r1-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129456,
          "output_tok_s": 381.58544791469023,
          "ttft_s": {
            "mean": 39.11297443829483,
            "p50": 36.25351455299824,
            "p90": 65.54042534709443,
            "p99": 66.34375637894759
          },
          "latency_s": {
            "mean": 77.7932884003892,
            "p50": 79.85938955300662,
            "p90": 100.19881022168265,
            "p99": 100.22422587619978
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 83.34489949242561,
          "ttft_s": {
            "mean": 16.724196845967526,
            "p50": 16.377652564493474,
            "p90": 21.860063897282817,
            "p99": 24.560182506071868
          },
          "latency_s": {
            "mean": 49.13745952684167,
            "p50": 49.1382106319943,
            "p90": 55.63348414349021,
            "p99": 55.63449692468246
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 7.127764236328813,
          "ttft_s": {
            "mean": 4.705665293127822,
            "p50": 0.2679350265098037,
            "p90": 14.551987214101244,
            "p99": 22.738563086616338
          },
          "latency_s": {
            "mean": 35.91542971751187,
            "p50": 31.847790159503347,
            "p90": 45.7358327610098,
            "p99": 54.14256675332115
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64677,
          "output_tok_s": 351.8727573605711,
          "ttft_s": {
            "mean": 3.521674936512113,
            "p50": 1.159026305496809,
            "p90": 12.42558311499306,
            "p99": 15.630398686900977
          },
          "latency_s": {
            "mean": 44.09513838454086,
            "p50": 44.52212094599963,
            "p90": 55.06858866651601,
            "p99": 55.075061435559476
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        }
      ]
    },
    {
      "name": "no-overlap",
      "state": "failed",
      "raw_completed": 0,
      "raw_failed": 0,
      "cuda_illegal_memory": true,
      "quality": {
        "structured_before.json": {
          "summary": {
            "json": "12/24",
            "tool": "0/8"
          },
          "raw_records_pass": false
        }
      },
      "medians": {},
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 1,
        "gpus": 4,
        "cuda_devices": "0,1,2,3",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 8192,
        "memory_fraction": 0.85,
        "prefix_caching": false,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": true,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 64,
        "engine_kv_tokens": 3343552,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": []
    },
    {
      "name": "state-pool-128",
      "state": "failed",
      "raw_completed": 41,
      "raw_failed": 64,
      "cuda_illegal_memory": true,
      "quality": {
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 1,
          "output_tok_s": 45.94293228252441,
          "ttft_p99_s": 14.285506313914665,
          "latency_p99_s": 16.416061545890404
        },
        "16": {
          "repeats": 1,
          "output_tok_s": 230.37919144087982,
          "ttft_p99_s": 13.079383591084042,
          "latency_p99_s": 18.95787175315054
        }
      },
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 1,
        "gpus": 4,
        "cuda_devices": "0,1,2,3",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 8192,
        "memory_fraction": 0.85,
        "prefix_caching": false,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 128,
        "engine_kv_tokens": 3157760,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": [
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 230.37919144087982,
          "ttft_s": {
            "mean": 11.536327121752038,
            "p50": 11.303974952519638,
            "p90": 12.264723458708612,
            "p99": 13.079383591084042
          },
          "latency_s": {
            "mean": 17.77264806803305,
            "p50": 17.772867571009556,
            "p90": 18.95620504229737,
            "p99": 18.95787175315054
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 45.94293228252441,
          "ttft_s": {
            "mean": 3.45167144361767,
            "p50": 0.19146403549530078,
            "p90": 11.751776408293516,
            "p99": 14.285506313914665
          },
          "latency_s": {
            "mean": 5.571619920494413,
            "p50": 2.3087469569873065,
            "p90": 13.875363507011206,
            "p99": 16.416061545890404
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "failed",
          "concurrency": 64,
          "requested": 256,
          "issued": 65,
          "completed": 1,
          "failed": 64,
          "output_tokens": 81,
          "output_tok_s": 1.104470235684331,
          "ttft_s": {
            "mean": 1.046222127013607,
            "p50": 1.046222127013607,
            "p90": 1.046222127013607,
            "p99": 1.046222127013607
          },
          "latency_s": {
            "mean": 4.9321645210147835,
            "p50": 4.9321645210147835,
            "p90": 4.9321645210147835,
            "p99": 4.9321645210147835
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true
          }
        }
      ]
    },
    {
      "name": "prefix-cache-on",
      "state": "passed",
      "raw_completed": 2424,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {
        "structured_after-first-sweep.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_after.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 3,
          "output_tok_s": 112.7721615234847,
          "ttft_p99_s": 0.15869988559483317,
          "latency_p99_s": 2.2925816984815173
        },
        "16": {
          "repeats": 3,
          "output_tok_s": 661.9492589294154,
          "ttft_p99_s": 0.36712518836808156,
          "latency_p99_s": 6.231433296954783
        },
        "64": {
          "repeats": 3,
          "output_tok_s": 1406.4431098498862,
          "ttft_p99_s": 3.3802405343536517,
          "latency_p99_s": 13.152764865489704
        },
        "128": {
          "repeats": 3,
          "output_tok_s": 1240.415277422261,
          "ttft_p99_s": 22.4791590929075,
          "latency_p99_s": 33.96179403249204
        }
      },
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 1,
        "gpus": 4,
        "cuda_devices": "0,1,2,3",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 8192,
        "memory_fraction": 0.85,
        "prefix_caching": true,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 575,
        "engine_kv_tokens": 1860480,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": [
        {
          "file": "r1-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129434,
          "output_tok_s": 889.2556382362172,
          "ttft_s": {
            "mean": 17.902615051418877,
            "p50": 12.724039923006785,
            "p90": 32.91898308222007,
            "p99": 35.28328892812103
          },
          "latency_s": {
            "mean": 34.53287414938052,
            "p50": 34.11946547801199,
            "p90": 53.94915376850404,
            "p99": 54.263225419922094
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 239.77806803516242,
          "ttft_s": {
            "mean": 11.198228920748988,
            "p50": 11.077324893994955,
            "p90": 11.621855959086679,
            "p99": 11.623322013140132
          },
          "latency_s": {
            "mean": 17.07730687543517,
            "p50": 17.076672781986417,
            "p90": 17.499832175610937,
            "p99": 17.501009427284007
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 31.274016157652184,
          "ttft_s": {
            "mean": 6.062830938124534,
            "p50": 5.9383444620034425,
            "p90": 11.945381194812944,
            "p99": 14.590837086483369
          },
          "latency_s": {
            "mean": 8.185282639995421,
            "p50": 8.059737960502389,
            "p90": 14.069022781809325,
            "p99": 16.721702152669312
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64734,
          "output_tok_s": 729.2034618962098,
          "ttft_s": {
            "mean": 3.839632379542195,
            "p50": 1.0031448229856323,
            "p90": 11.800510561995907,
            "p99": 11.974864140499268
          },
          "latency_s": {
            "mean": 21.921201689206328,
            "p50": 21.76161215301545,
            "p90": 32.64697553300357,
            "p99": 32.65606132200337
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129599,
          "output_tok_s": 1251.8624496441257,
          "ttft_s": {
            "mean": 13.095457514273846,
            "p50": 12.313535575507558,
            "p90": 21.627996011194774,
            "p99": 21.657545919224795
          },
          "latency_s": {
            "mean": 24.11957669959594,
            "p50": 23.507297032498172,
            "p90": 32.49556798368576,
            "p99": 32.520987741156425
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 662.12587413028,
          "ttft_s": {
            "mean": 0.3232818787510041,
            "p50": 0.31784115449409,
            "p90": 0.346356112594367,
            "p99": 0.34775608153548093
          },
          "latency_s": {
            "mean": 6.179577770218202,
            "p50": 6.180395362491254,
            "p90": 6.2146905788977165,
            "p99": 6.215481159040064
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 112.7721615234847,
          "ttft_s": {
            "mean": 0.14553320662889746,
            "p50": 0.14318407300743274,
            "p90": 0.15250045688881073,
            "p99": 0.15869988559483317
          },
          "latency_s": {
            "mean": 2.269609949868027,
            "p50": 2.266707848000806,
            "p90": 2.282100482986425,
            "p99": 2.2925816984815173
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64846,
          "output_tok_s": 1406.4431098498862,
          "ttft_s": {
            "mean": 0.7049455008087762,
            "p50": 0.658466076492914,
            "p90": 0.8704689709993545,
            "p99": 3.3777929882009627
          },
          "latency_s": {
            "mean": 11.231309496476342,
            "p50": 11.470050860996707,
            "p90": 11.564670274005039,
            "p99": 13.152764865489704
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129479,
          "output_tok_s": 1240.415277422261,
          "ttft_s": {
            "mean": 11.99741255822994,
            "p50": 12.167318688501837,
            "p90": 22.419171418680342,
            "p99": 22.4791590929075
          },
          "latency_s": {
            "mean": 24.234481008725936,
            "p50": 23.3892278179992,
            "p90": 33.94712836390245,
            "p99": 33.96179403249204
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 661.9492589294154,
          "ttft_s": {
            "mean": 0.3250069855021138,
            "p50": 0.32357691050856374,
            "p90": 0.36552545089507477,
            "p99": 0.36712518836808156
          },
          "latency_s": {
            "mean": 6.181522966594457,
            "p50": 6.18229974899441,
            "p90": 6.230435148303513,
            "p99": 6.231433296954783
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 112.78785258092671,
          "ttft_s": {
            "mean": 0.1447761861163599,
            "p50": 0.14282847849244718,
            "p90": 0.15124987699091436,
            "p99": 0.15738052668340968
          },
          "latency_s": {
            "mean": 2.269308627495775,
            "p50": 2.266360520006856,
            "p90": 2.277998927293811,
            "p99": 2.2913283797167243
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64685,
          "output_tok_s": 1411.032649334505,
          "ttft_s": {
            "mean": 0.6699291889523238,
            "p50": 0.5692388344905339,
            "p90": 0.8041502545092953,
            "p99": 3.3802405343536517
          },
          "latency_s": {
            "mean": 11.141236693539213,
            "p50": 11.381566248499439,
            "p90": 11.552835040492937,
            "p99": 13.119124967290524
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        }
      ]
    },
    {
      "name": "quad-ep4",
      "state": "passed",
      "raw_completed": 2424,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {
        "structured_after-first-sweep.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_after.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 3,
          "output_tok_s": 107.41192007845667,
          "ttft_p99_s": 0.16375609808164882,
          "latency_p99_s": 2.4030133180259146
        },
        "16": {
          "repeats": 3,
          "output_tok_s": 851.5517198406305,
          "ttft_p99_s": 0.5717285384223214,
          "latency_p99_s": 4.931252241091861
        },
        "64": {
          "repeats": 3,
          "output_tok_s": 1859.8673558548812,
          "ttft_p99_s": 2.9706487353119853,
          "latency_p99_s": 10.010227241709043
        },
        "128": {
          "repeats": 3,
          "output_tok_s": 1805.0300887899534,
          "ttft_p99_s": 11.796995312653307,
          "latency_p99_s": 20.558555910775613
        }
      },
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 4,
        "gpus": 4,
        "cuda_devices": "0,1,2,3",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 8192,
        "memory_fraction": 0.85,
        "prefix_caching": false,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 64,
        "engine_kv_tokens": 3342848,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": [
        {
          "file": "r1-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129557,
          "output_tok_s": 1260.6024251732163,
          "ttft_s": {
            "mean": 12.53803722177139,
            "p50": 10.110097048993339,
            "p90": 20.016680595217622,
            "p99": 21.580754566847734
          },
          "latency_s": {
            "mean": 24.303681303160772,
            "p50": 23.10884406400146,
            "p90": 37.7173691351898,
            "p99": 37.738446691651774
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 100.11255612767451,
          "ttft_s": {
            "mean": 36.4503651282821,
            "p50": 36.23815697801183,
            "p90": 62.26043774179707,
            "p99": 62.280563969950016
          },
          "latency_s": {
            "mean": 40.90760100159696,
            "p50": 40.90764927800046,
            "p90": 66.66657820190304,
            "p99": 66.66711389339441
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 45.21810391126722,
          "ttft_s": {
            "mean": 3.4339063963780063,
            "p50": 0.16165891451237258,
            "p90": 11.67099390651274,
            "p99": 14.294833029670698
          },
          "latency_s": {
            "mean": 5.660982587378385,
            "p50": 2.3858853394922335,
            "p90": 13.89953873140912,
            "p99": 16.5292428918535
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64856,
          "output_tok_s": 557.7702795009943,
          "ttft_s": {
            "mean": 7.946943641520079,
            "p50": 10.882383638003375,
            "p90": 11.639878339512506,
            "p99": 13.106567623553559
          },
          "latency_s": {
            "mean": 28.831674741941356,
            "p50": 29.00142634099757,
            "p90": 38.77019071349059,
            "p99": 40.16378140760641
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129601,
          "output_tok_s": 1805.0300887899534,
          "ttft_s": {
            "mean": 8.546234029683603,
            "p50": 9.621542494991445,
            "p90": 10.165536415114184,
            "p99": 10.737305122296673
          },
          "latency_s": {
            "mean": 16.584480028310224,
            "p50": 17.724076520011295,
            "p90": 18.341811590312865,
            "p99": 19.476030042565252
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 851.6334607915486,
          "ttft_s": {
            "mean": 0.44005207165537286,
            "p50": 0.32981640301295556,
            "p90": 0.5702070230938261,
            "p99": 0.5716393425539718
          },
          "latency_s": {
            "mean": 4.8029363406894845,
            "p50": 4.802801112498855,
            "p90": 4.9211950323981,
            "p99": 4.922980476900702
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 107.4501651660193,
          "ttft_s": {
            "mean": 0.15378701636655023,
            "p50": 0.1524830969865434,
            "p90": 0.15897070879873354,
            "p99": 0.16375609808164882
          },
          "latency_s": {
            "mean": 2.382087886242516,
            "p50": 2.381398363999324,
            "p90": 2.390479864287772,
            "p99": 2.4030133180259146
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64701,
          "output_tok_s": 1865.941037465073,
          "ttft_s": {
            "mean": 1.0181207390577356,
            "p50": 1.1340590879990486,
            "p90": 1.4621227134921355,
            "p99": 2.5654582946881384
          },
          "latency_s": {
            "mean": 8.456026786726056,
            "p50": 8.48716824798612,
            "p90": 8.849389845505357,
            "p99": 9.863487366345337
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129434,
          "output_tok_s": 1815.1680985296514,
          "ttft_s": {
            "mean": 8.514694302830605,
            "p50": 9.615311854504398,
            "p90": 10.052984977795859,
            "p99": 11.796995312653307
          },
          "latency_s": {
            "mean": 16.473991589518562,
            "p50": 17.790496807501768,
            "p90": 18.269239388717686,
            "p99": 20.558555910775613
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 851.5517198406305,
          "ttft_s": {
            "mean": 0.44057971434631327,
            "p50": 0.3294966515095439,
            "p90": 0.5702705850097118,
            "p99": 0.5717285384223214
          },
          "latency_s": {
            "mean": 4.803534745407887,
            "p50": 4.8039437955012545,
            "p90": 4.929219816499971,
            "p99": 4.931252241091861
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 107.41192007845667,
          "ttft_s": {
            "mean": 0.15332633862271905,
            "p50": 0.1528857139928732,
            "p90": 0.15569065380259417,
            "p99": 0.15953891177487095
          },
          "latency_s": {
            "mean": 2.3829073529923335,
            "p50": 2.37906252600078,
            "p90": 2.3911076866846996,
            "p99": 2.398212788266246
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64773,
          "output_tok_s": 1859.8673558548812,
          "ttft_s": {
            "mean": 0.9965048687029139,
            "p50": 0.8939575044933008,
            "p90": 1.411614193508285,
            "p99": 2.9706487353119853
          },
          "latency_s": {
            "mean": 8.459404966308057,
            "p50": 8.544932325501577,
            "p90": 8.805561418499565,
            "p99": 10.010227241709043
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        }
      ]
    },
    {
      "name": "quad-ep4-cache-state128",
      "state": "passed",
      "raw_completed": 2424,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {
        "structured_after-first-sweep.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_after.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 3,
          "output_tok_s": 106.747515945367,
          "ttft_p99_s": 0.17561265323427505,
          "latency_p99_s": 2.4157723497145343
        },
        "16": {
          "repeats": 3,
          "output_tok_s": 841.9392963922523,
          "ttft_p99_s": 0.4920894549929653,
          "latency_p99_s": 4.863305985671759
        },
        "64": {
          "repeats": 3,
          "output_tok_s": 1003.3864363753423,
          "ttft_p99_s": 12.94812130939972,
          "latency_p99_s": 18.692349989354263
        },
        "128": {
          "repeats": 3,
          "output_tok_s": 961.6962736121735,
          "ttft_p99_s": 29.258219619827287,
          "latency_p99_s": 35.65107141215616
        }
      },
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 4,
        "gpus": 4,
        "cuda_devices": "4,5,6,7",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 4096,
        "memory_fraction": 0.78,
        "prefix_caching": true,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 25,
        "effective_state_pool_slots": 128,
        "engine_kv_tokens": 2341824,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": [
        {
          "file": "r1-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129455,
          "output_tok_s": 956.7079944031192,
          "ttft_s": {
            "mean": 23.71343947704696,
            "p50": 26.900126177002676,
            "p90": 28.310075086183495,
            "p99": 29.214488517113494
          },
          "latency_s": {
            "mean": 29.877172266831906,
            "p50": 33.32296501100063,
            "p90": 34.689227238087916,
            "p99": 35.571237427328306
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 195.48985917184942,
          "ttft_s": {
            "mean": 16.101610908503062,
            "p50": 11.966962461505318,
            "p90": 21.117307949386305,
            "p99": 21.117886125578078
          },
          "latency_s": {
            "mean": 20.946435738782384,
            "p50": 20.9482620540075,
            "p90": 25.48898380229366,
            "p99": 25.489353399465617
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 36.9149383742273,
          "ttft_s": {
            "mean": 4.703812390878738,
            "p50": 0.9274880919983843,
            "p90": 11.67991373160039,
            "p99": 14.281941931565523
          },
          "latency_s": {
            "mean": 6.934514052121813,
            "p50": 3.160289396502776,
            "p90": 13.912161105091217,
            "p99": 16.52051802990725
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64743,
          "output_tok_s": 410.53959304591945,
          "ttft_s": {
            "mean": 24.351881201836022,
            "p50": 22.999608693498885,
            "p90": 39.3341525404976,
            "p99": 53.481013788592826
          },
          "latency_s": {
            "mean": 37.25119530923473,
            "p50": 36.59697410349327,
            "p90": 58.83053999900585,
            "p99": 59.27752779225848
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129612,
          "output_tok_s": 966.6504266330996,
          "ttft_s": {
            "mean": 23.55089713869421,
            "p50": 26.628633070999058,
            "p90": 28.44119509549055,
            "p99": 29.506082176380442
          },
          "latency_s": {
            "mean": 29.63690955136576,
            "p50": 32.81510124950728,
            "p90": 34.82588586789789,
            "p99": 35.98200612100511
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 846.497292693759,
          "ttft_s": {
            "mean": 0.46050222390476847,
            "p50": 0.46253862501180265,
            "p90": 0.4882421837130096,
            "p99": 0.4896041996995336
          },
          "latency_s": {
            "mean": 4.832631183344347,
            "p50": 4.8326217955036554,
            "p90": 4.851132637500996,
            "p99": 4.851899682886142
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 106.97008138009704,
          "ttft_s": {
            "mean": 0.1601547602476785,
            "p50": 0.15826401599042583,
            "p90": 0.1703671079070773,
            "p99": 0.1729699421796249
          },
          "latency_s": {
            "mean": 2.392839877997176,
            "p50": 2.3880720485030906,
            "p90": 2.407221384299919,
            "p99": 2.4157723497145343
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64672,
          "output_tok_s": 1003.3864363753423,
          "ttft_s": {
            "mean": 8.810769659179869,
            "p50": 8.653540580518893,
            "p90": 12.661787386983633,
            "p99": 12.802381099750345
          },
          "latency_s": {
            "mean": 14.368599323312537,
            "p50": 14.350411446983344,
            "p90": 18.417136695003137,
            "p99": 18.692349989354263
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129441,
          "output_tok_s": 961.6962736121735,
          "ttft_s": {
            "mean": 23.58657237420192,
            "p50": 26.81598853202013,
            "p90": 28.36445585679612,
            "p99": 29.258219619827287
          },
          "latency_s": {
            "mean": 29.70889271700031,
            "p50": 33.190224086501985,
            "p90": 34.698747103920326,
            "p99": 35.65107141215616
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 841.9392963922523,
          "ttft_s": {
            "mean": 0.4752180895948186,
            "p50": 0.48332261049654335,
            "p90": 0.49090519708697683,
            "p99": 0.4920894549929653
          },
          "latency_s": {
            "mean": 4.858949542408482,
            "p50": 4.858937267505098,
            "p90": 4.862438027211465,
            "p99": 4.863305985671759
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 106.747515945367,
          "ttft_s": {
            "mean": 0.16427642625421868,
            "p50": 0.1624820724973688,
            "p90": 0.17130656919907777,
            "p99": 0.17561265323427505
          },
          "latency_s": {
            "mean": 2.397859836128191,
            "p50": 2.394386927000596,
            "p90": 2.407943995296955,
            "p99": 2.411527713118121
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64732,
          "output_tok_s": 1003.685905764353,
          "ttft_s": {
            "mean": 8.876780010050425,
            "p50": 9.130291273992043,
            "p90": 12.770702822992462,
            "p99": 12.94812130939972
          },
          "latency_s": {
            "mean": 14.409709841804215,
            "p50": 14.649078597489279,
            "p90": 18.519582961002016,
            "p99": 18.668182650792005
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        }
      ]
    },
    {
      "name": "quad-ep4-mtp5",
      "state": "configuration_rejected",
      "raw_completed": 0,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {},
      "medians": {},
      "reason": "Startup reduced active request limit to51 because256 state slots/5 slots per request. Stopped before measurement; replaced with384 slots for intended64 active requests.",
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 4,
        "gpus": 4,
        "cuda_devices": "4,5,6,7",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 4096,
        "memory_fraction": 0.78,
        "prefix_caching": true,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 51,
        "effective_state_pool_slots": 256,
        "engine_kv_tokens": 789440,
        "spec_method": "EAGLE",
        "spec_num_steps": 5,
        "spec_num_tokens": 6
      },
      "points": []
    },
    {
      "name": "quad-ep4-cache-state384",
      "state": "passed",
      "raw_completed": 2424,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {
        "structured_after-first-sweep.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_after.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 3,
          "output_tok_s": 107.0263974111676,
          "ttft_p99_s": 0.1708983972138958,
          "latency_p99_s": 2.4144118319186965
        },
        "16": {
          "repeats": 3,
          "output_tok_s": 867.6706676816276,
          "ttft_p99_s": 0.3880503020714968,
          "latency_p99_s": 4.760133236583788
        },
        "64": {
          "repeats": 3,
          "output_tok_s": 1419.2578574234233,
          "ttft_p99_s": 7.233635237891512,
          "latency_p99_s": 19.24765919570491
        },
        "128": {
          "repeats": 3,
          "output_tok_s": 1813.6349010102242,
          "ttft_p99_s": 11.69733646986686,
          "latency_p99_s": 20.493336422392456
        }
      },
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 4,
        "gpus": 4,
        "cuda_devices": "4,5,6,7",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 4096,
        "memory_fraction": 0.78,
        "prefix_caching": true,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 384,
        "engine_kv_tokens": 1598848,
        "spec_method": null,
        "spec_num_steps": null,
        "spec_num_tokens": null
      },
      "points": [
        {
          "file": "r1-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129473,
          "output_tok_s": 1400.4675646056667,
          "ttft_s": {
            "mean": 12.34447461057465,
            "p50": 10.028282332015806,
            "p90": 19.812020549879527,
            "p99": 20.064431254085793
          },
          "latency_s": {
            "mean": 21.700847532111823,
            "p50": 18.873653044982348,
            "p90": 28.168505218790962,
            "p99": 28.177365138957104
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 196.02882049692124,
          "ttft_s": {
            "mean": 16.086940944747766,
            "p50": 11.715863708988763,
            "p90": 21.288617826689734,
            "p99": 21.289745817192305
          },
          "latency_s": {
            "mean": 20.887112958813304,
            "p50": 20.886942543496843,
            "p90": 25.660026548893075,
            "p99": 25.66092306558916
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 31.256066522270757,
          "ttft_s": {
            "mean": 5.958445636755641,
            "p50": 5.947007195005426,
            "p90": 11.66408836460323,
            "p99": 14.220015514467375
          },
          "latency_s": {
            "mean": 8.190060809629358,
            "p50": 8.178574143501464,
            "p90": 13.898369899613316,
            "p99": 16.461388405073084
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64759,
          "output_tok_s": 670.5786261724154,
          "ttft_s": {
            "mean": 4.6208277129895805,
            "p50": 1.424288265989162,
            "p90": 11.723520936007844,
            "p99": 21.66363114757114
          },
          "latency_s": {
            "mean": 23.914846341340763,
            "p50": 19.174328552995576,
            "p90": 29.074958990502637,
            "p99": 39.20626108659635
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129542,
          "output_tok_s": 1815.5174506841518,
          "ttft_s": {
            "mean": 8.449183242332651,
            "p50": 9.525558060980984,
            "p90": 9.983712433400797,
            "p99": 11.69733646986686
          },
          "latency_s": {
            "mean": 16.456383327889625,
            "p50": 17.784303518987144,
            "p90": 18.13120666638424,
            "p99": 20.493336422392456
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 867.6706676816276,
          "ttft_s": {
            "mean": 0.352469017245312,
            "p50": 0.35181698951055296,
            "p90": 0.38769192730542273,
            "p99": 0.3880503020714968
          },
          "latency_s": {
            "mean": 4.714448424433613,
            "p50": 4.715015337482328,
            "p90": 4.759411276513129,
            "p99": 4.760133236583788
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 107.09244676871378,
          "ttft_s": {
            "mean": 0.1564327863707149,
            "p50": 0.15427912799350452,
            "p90": 0.16377665419131518,
            "p99": 0.1708983972138958
          },
          "latency_s": {
            "mean": 2.390104033118405,
            "p50": 2.38538954698015,
            "p90": 2.4028260953869904,
            "p99": 2.4144118319186965
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64842,
          "output_tok_s": 1419.2578574234233,
          "ttft_s": {
            "mean": 1.1681433341175307,
            "p50": 0.9980013555032201,
            "p90": 1.6093061050050892,
            "p99": 7.233635237891512
          },
          "latency_s": {
            "mean": 11.184142195440586,
            "p50": 8.82843138849421,
            "p90": 19.23714463399665,
            "p99": 19.24765919570491
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129455,
          "output_tok_s": 1813.6349010102242,
          "ttft_s": {
            "mean": 8.422065902757822,
            "p50": 9.489459064003313,
            "p90": 9.922320897615283,
            "p99": 11.616353029200688
          },
          "latency_s": {
            "mean": 16.450615299468723,
            "p50": 17.583508135998272,
            "p90": 18.176266406077776,
            "p99": 20.177052196552975
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 874.828027932002,
          "ttft_s": {
            "mean": 0.3205911238110275,
            "p50": 0.3204320259974338,
            "p90": 0.32422791720018723,
            "p99": 0.3249581069918349
          },
          "latency_s": {
            "mean": 4.676273671027047,
            "p50": 4.677102997491602,
            "p90": 4.690455510301399,
            "p99": 4.691132902046665
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 107.0263974111676,
          "ttft_s": {
            "mean": 0.15722132300652447,
            "p50": 0.15594268900167663,
            "p90": 0.1636932496126974,
            "p99": 0.1681338688803953
          },
          "latency_s": {
            "mean": 2.3915936243793112,
            "p50": 2.388493604506948,
            "p90": 2.403896652592812,
            "p99": 2.4135086835647233
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64843,
          "output_tok_s": 1831.2371182309644,
          "ttft_s": {
            "mean": 1.0082383599650484,
            "p50": 0.9662621110037435,
            "p90": 1.580227320009726,
            "p99": 3.171599101647731
          },
          "latency_s": {
            "mean": 8.611651210363675,
            "p50": 8.697822767004254,
            "p90": 9.018755870507448,
            "p99": 10.295388958847584
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        }
      ]
    },
    {
      "name": "quad-ep4-mtp5-state384",
      "state": "passed",
      "raw_completed": 2424,
      "raw_failed": 0,
      "cuda_illegal_memory": false,
      "quality": {
        "structured_after-first-sweep.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_after.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        },
        "structured_before.json": {
          "summary": {
            "json": "24/24",
            "tool": "8/8"
          },
          "raw_records_pass": true
        }
      },
      "medians": {
        "1": {
          "repeats": 3,
          "output_tok_s": 208.7558173164237,
          "ttft_p99_s": 0.17400925119698513,
          "latency_p99_s": 1.5823841545614414
        },
        "16": {
          "repeats": 3,
          "output_tok_s": 1072.7883447798731,
          "ttft_p99_s": 0.4688684874289902,
          "latency_p99_s": 5.478851231299051
        },
        "64": {
          "repeats": 3,
          "output_tok_s": 1886.1493629615538,
          "ttft_p99_s": 1.4678033792981295,
          "latency_p99_s": 12.298147680924739
        },
        "128": {
          "repeats": 3,
          "output_tok_s": 1788.980407600423,
          "ttft_p99_s": 10.659513297432277,
          "latency_p99_s": 21.726272275114315
        }
      },
      "reason": null,
      "config": {
        "hf_repo": "zai-org/GLM-5.3-Flash",
        "model_revision": "eb9eb208eb0d988989d07a6a12d0fdeb5f52574a",
        "image_digest": "lmsysorg/sglang@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469",
        "engine_version": "v0.5.21",
        "tp": 4,
        "ep": 4,
        "gpus": 4,
        "cuda_devices": "4,5,6,7",
        "weight_dtype": "fp8",
        "kv_dtype": "bfloat16",
        "max_model_len": 32768,
        "max_num_batched_tokens": 4096,
        "memory_fraction": 0.78,
        "prefix_caching": true,
        "disable_cuda_graph": false,
        "disable_overlap_schedule": false,
        "requested_max_running_requests": 64,
        "effective_max_running_requests": 64,
        "effective_state_pool_slots": 384,
        "engine_kv_tokens": 241344,
        "spec_method": "EAGLE",
        "spec_num_steps": 5,
        "spec_num_tokens": 6
      },
      "points": [
        {
          "file": "r1-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129449,
          "output_tok_s": 1788.980407600423,
          "ttft_s": {
            "mean": 8.245502630505541,
            "p50": 9.35590585700993,
            "p90": 10.154320300617837,
            "p99": 10.659513297432277
          },
          "latency_s": {
            "mean": 16.672811292382676,
            "p50": 17.857692324512755,
            "p90": 20.166028014398762,
            "p99": 21.55309171098692
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 193.34899098118052,
          "ttft_s": {
            "mean": 7.05318876162346,
            "p50": 10.933310898995842,
            "p90": 12.468401548999829,
            "p99": 12.55913724155369
          },
          "latency_s": {
            "mean": 20.546258937314633,
            "p50": 20.858476362016518,
            "p90": 37.20453969119408,
            "p99": 38.13200308669475
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 56.08018948727163,
          "ttft_s": {
            "mean": 3.482475886492466,
            "p50": 0.17814820898638573,
            "p90": 11.784385720090357,
            "p99": 14.433722488011407
          },
          "latency_s": {
            "mean": 4.564499684373004,
            "p50": 1.4839298269944265,
            "p90": 12.998766866489312,
            "p99": 15.436449992348788
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r1-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64779,
          "output_tok_s": 545.7311534379046,
          "ttft_s": {
            "mean": 1.565135894125774,
            "p50": 0.2618781464843778,
            "p90": 1.8208711695042439,
            "p99": 11.967997844931961
          },
          "latency_s": {
            "mean": 28.902642609730947,
            "p50": 20.989729542008718,
            "p90": 68.77164141151297,
            "p99": 71.00313389726216
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129545,
          "output_tok_s": 1812.9991950030508,
          "ttft_s": {
            "mean": 8.317098436491108,
            "p50": 9.549717617497663,
            "p90": 10.273876459314488,
            "p99": 10.609479958447627
          },
          "latency_s": {
            "mean": 16.728587903405582,
            "p50": 18.261010443005944,
            "p90": 20.571858275981505,
            "p99": 21.726272275114315
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 1125.223427761395,
          "ttft_s": {
            "mean": 0.3114544975942408,
            "p50": 0.3577597980038263,
            "p90": 0.3821968297008425,
            "p99": 0.38309218089678326
          },
          "latency_s": {
            "mean": 3.42390501837599,
            "p50": 3.3871518335072324,
            "p90": 4.378169138703379,
            "p99": 4.959234003903403
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 215.96244823260085,
          "ttft_s": {
            "mean": 0.1651920104995952,
            "p50": 0.16577396250795573,
            "p90": 0.17011197450046894,
            "p99": 0.17320370263507356
          },
          "latency_s": {
            "mean": 1.1850425607481156,
            "p50": 1.1731745324941585,
            "p90": 1.3318966712948168,
            "p99": 1.4723205660234089
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r2-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64748,
          "output_tok_s": 1886.1493629615538,
          "ttft_s": {
            "mean": 0.45453259929365686,
            "p50": 0.25077419899753295,
            "p90": 1.196247776490054,
            "p99": 1.383654780741199
          },
          "latency_s": {
            "mean": 8.177411416445011,
            "p50": 8.43034201048431,
            "p90": 10.920385110002826,
            "p99": 12.298147680924739
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c128_summary.json",
          "state": "passed",
          "concurrency": 128,
          "requested": 512,
          "issued": 512,
          "completed": 512,
          "failed": 0,
          "output_tokens": 129525,
          "output_tok_s": 1781.9556597609305,
          "ttft_s": {
            "mean": 8.416077384866696,
            "p50": 9.590674457504065,
            "p90": 10.230138231505407,
            "p99": 10.887093645761487
          },
          "latency_s": {
            "mean": 16.953964888145094,
            "p50": 18.19422331849637,
            "p90": 20.654836634595995,
            "p99": 21.978145165664028
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c16_summary.json",
          "state": "passed",
          "concurrency": 16,
          "requested": 32,
          "issued": 32,
          "completed": 32,
          "failed": 0,
          "output_tokens": 8192,
          "output_tok_s": 1072.7883447798731,
          "ttft_s": {
            "mean": 0.32959206062605517,
            "p50": 0.33666530648770276,
            "p90": 0.44550928859971484,
            "p99": 0.4688684874289902
          },
          "latency_s": {
            "mean": 3.4865789926889192,
            "p50": 3.295975904504303,
            "p90": 4.552401547500631,
            "p99": 5.478851231299051
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c1_summary.json",
          "state": "passed",
          "concurrency": 1,
          "requested": 8,
          "issued": 8,
          "completed": 8,
          "failed": 0,
          "output_tokens": 2048,
          "output_tok_s": 208.7558173164237,
          "ttft_s": {
            "mean": 0.16650415562617127,
            "p50": 0.16590189601993188,
            "p90": 0.17356520199682562,
            "p99": 0.17400925119698513
          },
          "latency_s": {
            "mean": 1.2259577072509273,
            "p50": 1.1743361554836156,
            "p90": 1.5146410244051367,
            "p99": 1.5823841545614414
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        },
        {
          "file": "r3-c64_summary.json",
          "state": "passed",
          "concurrency": 64,
          "requested": 256,
          "issued": 256,
          "completed": 256,
          "failed": 0,
          "output_tokens": 64757,
          "output_tok_s": 1928.7288316906258,
          "ttft_s": {
            "mean": 0.4592012473358409,
            "p50": 0.25204493549244944,
            "p90": 1.0024924500030465,
            "p99": 1.4678033792981295
          },
          "latency_s": {
            "mean": 8.010479643573944,
            "p50": 8.287515199495829,
            "p90": 10.624594943001284,
            "p99": 12.206432534151826
          },
          "audit": {
            "raw_count": true,
            "completed": true,
            "failed": true,
            "tokens": true,
            "all_requested": true
          }
        }
      ]
    }
  ]
}