{
  "schema": "anvil-serving.benchmark-graph-manifest/v1",
  "title": "GLM Flare whole-recipe scout",
  "subtitle": "Different checkpoints; variable output; n8/cell; C4 warm after C1. No promotion.",
  "output": "graphs/matrix.svg",
  "data_output": "graphs/graph-data.json",
  "metric_semantics": {
    "tpot": "Per-request aggregate proxy; not raw token interval distribution.",
    "prefill": "Includes queueing, scheduling, prefill and first-token work.",
    "throughput": "Variable output lengths; descriptive scout only."
  },
  "charts": [
    {
      "title": "Decode: C1 near-zero hits",
      "metric": "metrics.decode_tok_s_p50",
      "x_label": "Actual mean prompt tokens",
      "y_label": "tokens/s",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Aggregate output: C1 near-zero hits",
      "metric": "metrics.throughput_tok_s",
      "x_label": "Actual mean prompt tokens",
      "y_label": "tokens/s",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "First any output: C1 near-zero hits",
      "metric": "metrics.time_to_first_output_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Visible TTFT: C1 near-zero hits",
      "metric": "metrics.ttft_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "TPOT proxy: C1 near-zero hits",
      "metric": "metrics.tpot_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms/token",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "End to end: C1 near-zero hits",
      "metric": "metrics.e2e_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Effective prefill: C1 near-zero hits",
      "metric": "metrics.effective_prefill_tok_s_p50",
      "x_label": "Actual mean prompt tokens",
      "y_label": "tokens/s",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c1.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c1.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c1.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Decode: C4 warm",
      "metric": "metrics.decode_tok_s_p50",
      "x_label": "Actual mean prompt tokens",
      "y_label": "tokens/s",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Aggregate output: C4 warm",
      "metric": "metrics.throughput_tok_s",
      "x_label": "Actual mean prompt tokens",
      "y_label": "tokens/s",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "First any output: C4 warm",
      "metric": "metrics.time_to_first_output_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Visible TTFT: C4 warm",
      "metric": "metrics.ttft_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "TPOT proxy: C4 warm",
      "metric": "metrics.tpot_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms/token",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "End to end: C4 warm",
      "metric": "metrics.e2e_p50_ms",
      "x_label": "Actual mean prompt tokens",
      "y_label": "ms",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Effective prefill: C4 warm",
      "metric": "metrics.effective_prefill_tok_s_p50",
      "x_label": "Actual mean prompt tokens",
      "y_label": "tokens/s",
      "series": [
        {
          "label": "Flare QAD",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/flare-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/flare-natural-ctx32768-c4.json"
            }
          ]
        },
        {
          "label": "Incumbent FP4",
          "points": [
            {
              "x": "7.7K",
              "artifact": "raw/incumbent-natural-ctx10240-c4.json"
            },
            {
              "x": "22.9K",
              "artifact": "raw/incumbent-natural-ctx32768-c4.json"
            }
          ]
        }
      ]
    }
  ]
}