maas-base/docker-compose/grafana/grafana_dashboards/gpustack-model.json

{
  "annotations": {
    "list": [
      {
        "builtIn": 1,
        "datasource": {
          "type": "grafana",
          "uid": "-- Grafana --"
        },
        "enable": true,
        "hide": true,
        "iconColor": "rgba(0, 211, 255, 1)",
        "name": "Annotations & Alerts",
        "target": {
          "limit": 100,
          "matchAny": false,
          "tags": [],
          "type": "dashboard"
        },
        "type": "dashboard"
      }
    ]
  },
  "description": "Monitoring Inference Server",
  "editable": true,
  "fiscalYearStartMonth": 0,
  "graphTooltip": 0,
  "id": 0,
  "links": [],
  "panels": [
    {
      "collapsed": false,
      "gridPos": {
        "h": 1,
        "w": 24,
        "x": 0,
        "y": 0
      },
      "id": 24,
      "panels": [],
      "title": "Summary",
      "type": "row"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "max": 90,
          "min": 0,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "#EAB839",
                "value": 60
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "none"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 3,
        "w": 4,
        "x": 0,
        "y": 1
      },
      "id": 32,
      "options": {
        "colorMode": "none",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "horizontal",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "showPercentChange": false,
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "P1809F7CD0C75ACF3"
          },
          "editorMode": "code",
          "expr": "sum(gpustack:model_desired_instances{cluster_name=\"$cluster_name\", model_name=\"$model_name\"})",
          "instant": false,
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Desired Instances",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "max": 90,
          "min": 0,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "#EAB839",
                "value": 60
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "none"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 3,
        "w": 4,
        "x": 4,
        "y": 1
      },
      "id": 33,
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "horizontal",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "showPercentChange": false,
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "P1809F7CD0C75ACF3"
          },
          "editorMode": "code",
          "expr": "sum(gpustack:model_running_instances{cluster_name=\"$cluster_name\",model_name=\"$model_name\"})",
          "instant": false,
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Running Instances",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "max": 90,
          "min": 0,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "#EAB839",
                "value": 60
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "none"
        },
        "overrides": [
          {
            "matcher": {
              "id": "byName",
              "options": "Value"
            },
            "properties": [
              {
                "id": "color",
                "value": {
                  "mode": "continuous-RdYlGr"
                }
              }
            ]
          }
        ]
      },
      "gridPos": {
        "h": 3,
        "w": 4,
        "x": 8,
        "y": 1
      },
      "id": 35,
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "horizontal",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "showPercentChange": false,
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "P1809F7CD0C75ACF3"
          },
          "editorMode": "code",
          "expr": "sum(gpustack:model_desired_instances{cluster_name=\"$cluster_name\",model_name=\"$model_name\"}) - sum(gpustack:model_running_instances{cluster_name=\"$cluster_name\",model_name=\"$model_name\"})",
          "instant": false,
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Error Instances",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "max": 90,
          "min": 0,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "#EAB839",
                "value": 60
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "none"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 3,
        "w": 4,
        "x": 12,
        "y": 1
      },
      "id": 37,
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "horizontal",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "showPercentChange": false,
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "P1809F7CD0C75ACF3"
          },
          "editorMode": "code",
          "expr": "sum(gpustack:num_requests_running{cluster_name=\"$cluster_name\",model_name=\"$model_name\"})",
          "instant": false,
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Running Requests",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "max": 90,
          "min": 0,
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "#EAB839",
                "value": 60
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "none"
        },
        "overrides": [
          {
            "matcher": {
              "id": "byName",
              "options": "Value"
            },
            "properties": [
              {
                "id": "color",
                "value": {
                  "mode": "continuous-RdYlGr"
                }
              }
            ]
          }
        ]
      },
      "gridPos": {
        "h": 3,
        "w": 4,
        "x": 16,
        "y": 1
      },
      "id": 36,
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "horizontal",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "",
          "values": false
        },
        "showPercentChange": false,
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "P1809F7CD0C75ACF3"
          },
          "editorMode": "code",
          "expr": "sum(gpustack:num_requests_waiting{cluster_name=\"$cluster_name\",model_name=\"$model_name\"})",
          "instant": false,
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Waiting Requests",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic-by-name"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 3,
        "w": 4,
        "x": 20,
        "y": 1
      },
      "id": 28,
      "options": {
        "colorMode": "value",
        "graphMode": "area",
        "justifyMode": "center",
        "orientation": "auto",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "/^runtime$/",
          "values": false
        },
        "showPercentChange": false,
        "text": {},
        "textMode": "value",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "editorMode": "code",
          "exemplar": false,
          "expr": "gpustack:model_info{cluster_name=\"$cluster_name\",model_name=\"$model_name\"}",
          "format": "table",
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Runtime",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 4,
        "w": 12,
        "x": 0,
        "y": 4
      },
      "id": 34,
      "options": {
        "colorMode": "background",
        "graphMode": "none",
        "justifyMode": "auto",
        "orientation": "auto",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "/^model_name$/",
          "values": false
        },
        "showPercentChange": false,
        "text": {
          "valueSize": 55
        },
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "editorMode": "code",
          "exemplar": false,
          "expr": "gpustack:model_info{cluster_name=\"$cluster_name\",model_name=\"$model_name\"}",
          "format": "table",
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Model",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": [
          {
            "matcher": {
              "id": "byName",
              "options": "source_key"
            },
            "properties": [
              {
                "id": "color",
                "value": {
                  "mode": "palette-classic-by-name"
                }
              }
            ]
          }
        ]
      },
      "gridPos": {
        "h": 4,
        "w": 8,
        "x": 12,
        "y": 4
      },
      "id": 30,
      "options": {
        "colorMode": "background_solid",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "auto",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "/^source_key$/",
          "values": false
        },
        "showPercentChange": false,
        "text": {
          "valueSize": 30
        },
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "editorMode": "code",
          "exemplar": false,
          "expr": "gpustack:model_info{cluster_name=\"$cluster_name\",model_name=\"$model_name\"}",
          "format": "table",
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Model Source Name",
      "type": "stat"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "thresholds"
          },
          "mappings": [
            {
              "options": {
                "huggingface": {
                  "index": 0,
                  "text": "Hugging Face"
                },
                "local_path": {
                  "index": 2,
                  "text": "Local Path"
                },
                "model_scope": {
                  "index": 1,
                  "text": "ModelScope"
                }
              },
              "type": "value"
            }
          ],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": [
          {
            "matcher": {
              "id": "byName",
              "options": "source"
            },
            "properties": [
              {
                "id": "color",
                "value": {
                  "mode": "palette-classic-by-name"
                }
              }
            ]
          }
        ]
      },
      "gridPos": {
        "h": 4,
        "w": 4,
        "x": 20,
        "y": 4
      },
      "id": 29,
      "options": {
        "colorMode": "background",
        "graphMode": "area",
        "justifyMode": "auto",
        "orientation": "auto",
        "percentChangeColorMode": "standard",
        "reduceOptions": {
          "calcs": [
            "lastNotNull"
          ],
          "fields": "/^source$/",
          "values": false
        },
        "showPercentChange": false,
        "text": {
          "valueSize": 30
        },
        "textMode": "auto",
        "wideLayout": true
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "editorMode": "code",
          "exemplar": false,
          "expr": "gpustack:model_info{cluster_name=\"$cluster_name\",model_name=\"$model_name\"}",
          "format": "table",
          "legendFormat": "__auto",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Model Source",
      "type": "stat"
    },
    {
      "collapsed": false,
      "gridPos": {
        "h": 1,
        "w": 24,
        "x": 0,
        "y": 8
      },
      "id": 23,
      "panels": [],
      "title": "Service API",
      "type": "row"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Total request per minute.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "req/min"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 9
      },
      "id": 22,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "editorMode": "code",
          "expr": "increase(gpustack:e2e_request_latency_seconds_count{cluster_name=\"$cluster_name\",model_name=~\"$model_name\", model_instance_name=~\"$model_instance_name\"}[1m])",
          "legendFormat": "Requests/min({{model_instance_name}})",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Total Requests per Minute",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Number of tokens generation per second for current model",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 9
      },
      "id": 41,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum(rate(gpustack:generation_tokens_total{cluster_name=\"$cluster_name\",model_name=\"$model_name\"}[$__rate_interval]))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Generation Tokens/Sec (Total) ",
          "range": true,
          "refId": "D",
          "useBackend": false
        }
      ],
      "title": "Model Token Generation Throughput",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "End to end request latency measured in seconds.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 17
      },
      "id": 9,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:e2e_request_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:e2e_request_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:e2e_request_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:e2e_request_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:e2e_request_latency_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:e2e_request_latency_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Average({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": "End-to-End Request Latency",
      "type": "timeseries"
    },
    {
      "datasource": {
        "default": true,
        "type": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "custom": {
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "scaleDistribution": {
              "type": "linear"
            }
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 17
      },
      "id": 40,
      "maxDataPoints": 30,
      "options": {
        "calculate": false,
        "calculation": {
          "yBuckets": {
            "scale": {
              "type": "linear"
            }
          }
        },
        "cellGap": 1,
        "cellValues": {},
        "color": {
          "exponent": 0.5,
          "fill": "dark-orange",
          "mode": "scheme",
          "reverse": false,
          "scale": "exponential",
          "scheme": "Spectral",
          "steps": 64
        },
        "exemplars": {
          "color": "rgba(255,0,255,0.7)"
        },
        "filterValues": {
          "le": 1e-09
        },
        "legend": {
          "show": true
        },
        "rowsFrame": {
          "layout": "auto"
        },
        "tooltip": {
          "mode": "single",
          "showColorScale": true,
          "yHistogram": false
        },
        "yAxis": {
          "axisPlacement": "left",
          "reverse": false,
          "unit": "secs"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum by(le) (increase(gpustack:e2e_request_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=~\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]))",
          "format": "heatmap",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{le}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        }
      ],
      "title": "End-to-End Request Latency Heatmap",
      "type": "heatmap"
    },
    {
      "collapsed": false,
      "gridPos": {
        "h": 1,
        "w": 24,
        "x": 0,
        "y": 25
      },
      "id": 25,
      "panels": [],
      "title": "Runtime Performance",
      "type": "row"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Number of requests in RUNNING, WAITING, and SWAPPED state",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "none"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 26
      },
      "id": 3,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "gpustack:num_requests_running{cluster_name=\"$cluster_name\",model_instance_name=~\"$model_instance_name\", model_name=\"$model_name\"}",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "Request Running({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "gpustack:num_requests_swapped{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "Request Swapped({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "gpustack:num_requests_waiting{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "Request Waiting({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        }
      ],
      "title": "Request Status",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Total number of tokens processed per second, combining prompt and generation tokens.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 26
      },
      "id": 8,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "rate(gpustack:prompt_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]) + rate(gpustack:generation_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Total Tokens/Sec {{model_instance_name}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "sum(rate(gpustack:prompt_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]) + rate(gpustack:generation_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Total Tokens/Sec (Total)",
          "range": true,
          "refId": "B",
          "useBackend": false
        }
      ],
      "title": "Token Throughput",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "P50, P90, P95, and P99 TTFT latency in seconds.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 42
      },
      "id": 5,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:time_to_first_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:time_to_first_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:time_to_first_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:time_to_first_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:time_to_first_token_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:time_to_first_token_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Average({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": "Time To First Token Latency",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Inter token latency in seconds.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 42
      },
      "id": 39,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:inter_token_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:inter_token_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:inter_token_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:inter_token_latency_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:inter_token_latency_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:inter_token_latency_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Mean({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": " Inter Token Latency",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Percentage of used cache blocks by LLM Serving Runtime.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "percentunit"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 50
      },
      "id": 4,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "gpustack:kv_cache_usage_ratio{cluster_name=\"$cluster_name\",model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}",
          "instant": false,
          "legendFormat": "{{model_instance_name}}",
          "range": true,
          "refId": "A"
        }
      ],
      "title": "Cache Utilization",
      "type": "timeseries"
    },
    {
      "datasource": {
        "default": true,
        "type": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "percentunit"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 50
      },
      "id": 38,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "gpustack:prefix_cache_hit_rate{model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{model_instance_name}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:prefix_cache_hits_total{cluster_name=\"$cluster_name\",model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}[1m]) / rate(gpustack:prefix_cache_queries_total{cluster_name=\"$cluster_name\",model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}[1m])",
          "hide": false,
          "instant": false,
          "legendFormat": "{{model_instance_name}}",
          "range": true,
          "refId": "B"
        }
      ],
      "title": "Cache Hit Rate",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Heatmap of request prompt length",
      "fieldConfig": {
        "defaults": {
          "custom": {
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "scaleDistribution": {
              "type": "linear"
            }
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 58
      },
      "id": 12,
      "options": {
        "calculate": false,
        "cellGap": 1,
        "cellValues": {
          "unit": "none"
        },
        "color": {
          "exponent": 0.5,
          "fill": "dark-orange",
          "min": 0,
          "mode": "scheme",
          "reverse": false,
          "scale": "exponential",
          "scheme": "Spectral",
          "steps": 64
        },
        "exemplars": {
          "color": "rgba(255,0,255,0.7)"
        },
        "filterValues": {
          "le": 1e-09
        },
        "legend": {
          "show": true
        },
        "rowsFrame": {
          "layout": "auto",
          "value": "Request count"
        },
        "tooltip": {
          "mode": "single",
          "showColorScale": false,
          "yHistogram": true
        },
        "yAxis": {
          "axisLabel": "Prompt Length",
          "axisPlacement": "left",
          "reverse": false,
          "unit": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum by(le) (increase(gpustack:request_prompt_tokens_bucket{model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\", cluster_name=\"$cluster_name\"}[$__rate_interval]))",
          "format": "heatmap",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{le}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        }
      ],
      "title": "Request Prompt Length",
      "type": "heatmap"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Heatmap of request generation length",
      "fieldConfig": {
        "defaults": {
          "custom": {
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "scaleDistribution": {
              "type": "linear"
            }
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 58
      },
      "id": 13,
      "options": {
        "calculate": false,
        "cellGap": 1,
        "cellValues": {
          "unit": "none"
        },
        "color": {
          "exponent": 0.5,
          "fill": "dark-orange",
          "min": 0,
          "mode": "scheme",
          "reverse": false,
          "scale": "exponential",
          "scheme": "Spectral",
          "steps": 64
        },
        "exemplars": {
          "color": "rgba(255,0,255,0.7)"
        },
        "filterValues": {
          "le": 1e-09
        },
        "legend": {
          "show": true
        },
        "rowsFrame": {
          "layout": "auto",
          "value": "Request count"
        },
        "tooltip": {
          "mode": "single",
          "showColorScale": false,
          "yHistogram": true
        },
        "yAxis": {
          "axisLabel": "Generation Length",
          "axisPlacement": "left",
          "reverse": false,
          "unit": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum by(le) (increase(gpustack:request_generation_tokens_bucket{model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\", cluster_name=\"$cluster_name\"}[$__rate_interval]))",
          "format": "heatmap",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{le}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        }
      ],
      "title": "Request Generation Length",
      "type": "heatmap"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Per output token latency in seconds.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 66
      },
      "id": 10,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:time_per_output_token_seconds_bucket{model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\", cluster_name=\"$cluster_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:time_per_output_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:time_per_output_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:time_per_output_token_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:time_per_output_token_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:time_per_output_token_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Mean({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": "Time Per Output Token Latency",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Number of finished requests by their finish reason: either an EOS token was generated or the max sequence length was reached.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 66
      },
      "id": 11,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum by(finished_reason, model_instance_name) (increase(gpustack:request_success_total{model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\", cluster_name=\"$cluster_name\"}[$__rate_interval]))",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "interval": "",
          "legendFormat": "{{finished_reason}}({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        }
      ],
      "title": "Finish Reason",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Number of prompt tokens and cached prompt tokens served per second.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 34
      },
      "id": 42,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "rate(gpustack:prompt_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Prompt Tokens/Sec {{model_instance_name}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "rate(gpustack:prompt_tokens_cached_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Cached Prompt Tokens/Sec {{model_instance_name}}",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "sum(rate(gpustack:prompt_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Prompt Tokens/Sec (Total)",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "sum(rate(gpustack:prompt_tokens_cached_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Cached Prompt Tokens/Sec (Total)",
          "range": true,
          "refId": "D",
          "useBackend": false
        }
      ],
      "title": "Prompt Tokens Throughput",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Time spent waiting in the request queue.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 74
      },
      "id": 43,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:request_queue_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:request_queue_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:request_queue_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:request_queue_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:request_queue_time_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:request_queue_time_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Mean({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": "Request Queue Time",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Time spent in the prefill phase.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 74
      },
      "id": 44,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:request_prefill_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:request_prefill_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:request_prefill_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:request_prefill_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:request_prefill_time_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:request_prefill_time_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Mean({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": "Request Prefill Time",
      "type": "timeseries"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Time spent in the decode phase.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "s"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 82
      },
      "id": 45,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.99, sum by(le, model_instance_name) (rate(gpustack:request_decode_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P99({{model_instance_name}})",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.95, sum by(le, model_instance_name) (rate(gpustack:request_decode_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P95({{model_instance_name}})",
          "range": true,
          "refId": "B",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.9, sum by(le, model_instance_name) (rate(gpustack:request_decode_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P90({{model_instance_name}})",
          "range": true,
          "refId": "C",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "histogram_quantile(0.5, sum by(le, model_instance_name) (rate(gpustack:request_decode_time_seconds_bucket{cluster_name=\"$cluster_name\",model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "P50({{model_instance_name}})",
          "range": true,
          "refId": "D",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "editorMode": "code",
          "expr": "rate(gpustack:request_decode_time_seconds_sum{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])\n/\nrate(gpustack:request_decode_time_seconds_count{cluster_name=\"$cluster_name\",model_name=\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "hide": false,
          "instant": false,
          "legendFormat": "Mean({{model_instance_name}})",
          "range": true,
          "refId": "E"
        }
      ],
      "title": "Request Decode Time",
      "type": "timeseries"
    },
    {
      "datasource": {
        "default": true,
        "type": "prometheus"
      },
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          },
          "unit": "percentunit"
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 82
      },
      "id": 46,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "gpustack:spec_decode_accept_rate{cluster_name=\"$cluster_name\",model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{model_instance_name}} (SGLang)",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "rate(gpustack:spec_decode_accepted_tokens_total{cluster_name=\"$cluster_name\",model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval]) / rate(gpustack:spec_decode_draft_tokens_total{cluster_name=\"$cluster_name\",model_name=~\"$model_name\",model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{model_instance_name}} (vLLM)",
          "range": true,
          "refId": "B",
          "useBackend": false
        }
      ],
      "title": "Spec Decode Accept Rate",
      "type": "timeseries",
      "description": "Average speculative decoding acceptance rate."
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Heatmap of requested number of parallel completions per request.",
      "fieldConfig": {
        "defaults": {
          "custom": {
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "scaleDistribution": {
              "type": "linear"
            }
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 90
      },
      "id": 48,
      "options": {
        "calculate": false,
        "cellGap": 1,
        "cellValues": {
          "unit": "none"
        },
        "color": {
          "exponent": 0.5,
          "fill": "dark-orange",
          "min": 0,
          "mode": "scheme",
          "reverse": false,
          "scale": "exponential",
          "scheme": "Spectral",
          "steps": 64
        },
        "exemplars": {
          "color": "rgba(255,0,255,0.7)"
        },
        "filterValues": {
          "le": 1e-09
        },
        "legend": {
          "show": true
        },
        "rowsFrame": {
          "layout": "auto",
          "value": "Request count"
        },
        "tooltip": {
          "mode": "single",
          "showColorScale": false,
          "yHistogram": true
        },
        "yAxis": {
          "axisLabel": "Requested N",
          "axisPlacement": "left",
          "reverse": false,
          "unit": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum by(le) (increase(gpustack:request_params_n_bucket{model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\", cluster_name=\"$cluster_name\"}[$__rate_interval]))",
          "format": "heatmap",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{le}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        }
      ],
      "title": "Request Params N",
      "type": "heatmap"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Heatmap of requested max_tokens per request.",
      "fieldConfig": {
        "defaults": {
          "custom": {
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "scaleDistribution": {
              "type": "linear"
            }
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 90
      },
      "id": 49,
      "options": {
        "calculate": false,
        "cellGap": 1,
        "cellValues": {
          "unit": "none"
        },
        "color": {
          "exponent": 0.5,
          "fill": "dark-orange",
          "min": 0,
          "mode": "scheme",
          "reverse": false,
          "scale": "exponential",
          "scheme": "Spectral",
          "steps": 64
        },
        "exemplars": {
          "color": "rgba(255,0,255,0.7)"
        },
        "filterValues": {
          "le": 1e-09
        },
        "legend": {
          "show": true
        },
        "rowsFrame": {
          "layout": "auto",
          "value": "Request count"
        },
        "tooltip": {
          "mode": "single",
          "showColorScale": false,
          "yHistogram": true
        },
        "yAxis": {
          "axisLabel": "Requested Max Tokens",
          "axisPlacement": "left",
          "reverse": false,
          "unit": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "builder",
          "expr": "sum by(le) (increase(gpustack:request_params_max_tokens_bucket{model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\", cluster_name=\"$cluster_name\"}[$__rate_interval]))",
          "format": "heatmap",
          "fullMetaSearch": false,
          "includeNullMetadata": true,
          "instant": false,
          "legendFormat": "{{le}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        }
      ],
      "title": "Request Max Tokens",
      "type": "heatmap"
    },
    {
      "datasource": {
        "type": "prometheus",
        "uid": "prometheus"
      },
      "description": "Number of generation tokens processed per second.",
      "fieldConfig": {
        "defaults": {
          "color": {
            "mode": "palette-classic"
          },
          "custom": {
            "axisBorderShow": false,
            "axisCenteredZero": false,
            "axisColorMode": "text",
            "axisLabel": "",
            "axisPlacement": "auto",
            "barAlignment": 0,
            "barWidthFactor": 0.6,
            "drawStyle": "line",
            "fillOpacity": 0,
            "gradientMode": "none",
            "hideFrom": {
              "legend": false,
              "tooltip": false,
              "viz": false
            },
            "insertNulls": false,
            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "showValues": false,
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }
          },
          "mappings": [],
          "thresholds": {
            "mode": "absolute",
            "steps": [
              {
                "color": "green",
                "value": 0
              },
              {
                "color": "red",
                "value": 80
              }
            ]
          }
        },
        "overrides": []
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 34
      },
      "id": 47,
      "options": {
        "legend": {
          "calcs": [
            "last",
            "max"
          ],
          "displayMode": "table",
          "placement": "bottom",
          "showLegend": true
        },
        "tooltip": {
          "hideZeros": false,
          "mode": "single",
          "sort": "none"
        }
      },
      "pluginVersion": "12.2.0",
      "targets": [
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "rate(gpustack:generation_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval])",
          "fullMetaSearch": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Generation Tokens/Sec {{model_instance_name}}",
          "range": true,
          "refId": "A",
          "useBackend": false
        },
        {
          "datasource": {
            "type": "prometheus",
            "uid": "Prometheus"
          },
          "disableTextWrap": false,
          "editorMode": "code",
          "expr": "sum(rate(gpustack:generation_tokens_total{cluster_name=\"$cluster_name\", model_name=\"$model_name\", model_instance_name=~\"$model_instance_name\"}[$__rate_interval]))",
          "fullMetaSearch": false,
          "hide": false,
          "includeNullMetadata": false,
          "instant": false,
          "legendFormat": "Generation Tokens/Sec (Total)",
          "range": true,
          "refId": "B",
          "useBackend": false
        }
      ],
      "title": "Generation Tokens Throughput",
      "type": "timeseries"
    }
  ],
  "preload": false,
  "refresh": "",
  "schemaVersion": 42,
  "tags": [],
  "templating": {
    "list": [
      {
        "current": {
          "text": "",
          "value": ""
        },
        "definition": "query_result(gpustack:cluster_info)",
        "label": "Cluster",
        "name": "cluster_name",
        "options": [],
        "query": {
          "qryType": 1,
          "query": "query_result(gpustack:cluster_info)",
          "refId": "PrometheusVariableQueryEditor-VariableQuery"
        },
        "refresh": 2,
        "regex": "/cluster_name=\"([^\"]+)\"/",
        "type": "query"
      },
      {
        "current": {
          "text": "",
          "value": ""
        },
        "definition": "query_result(gpustack:model_info{cluster_name=\"$cluster_name\"})",
        "label": "Model",
        "name": "model_name",
        "options": [],
        "query": {
          "qryType": 1,
          "query": "query_result(gpustack:model_info{cluster_name=\"$cluster_name\"})",
          "refId": "PrometheusVariableQueryEditor-VariableQuery"
        },
        "refresh": 2,
        "regex": "/model_name=\"([^\"]+)\"/",
        "type": "query"
      },
      {
        "allValue": ".*",
        "current": {
          "text": "All",
          "value": "$__all"
        },
        "definition": "query_result(gpustack:model_instance_status{model_name=\"$model_name\", cluster_name=\"$cluster_name\"})",
        "includeAll": true,
        "label": "Model Instance",
        "name": "model_instance_name",
        "options": [],
        "query": {
          "qryType": 1,
          "query": "query_result(gpustack:model_instance_status{model_name=\"$model_name\", cluster_name=\"$cluster_name\"})",
          "refId": "PrometheusVariableQueryEditor-VariableQuery"
        },
        "refresh": 2,
        "regex": "/model_instance_name=\"([^\"]+)\"/",
        "type": "query"
      }
    ]
  },
  "time": {
    "from": "now-5m",
    "to": "now"
  },
  "timepicker": {},
  "timezone": "",
  "title": "GPUStack Model",
  "uid": "gpustack-model",
  "version": 2
}