Upload shared/metrics-schema.json with huggingface_hub

shared/metrics-schema.json

@@ -0,0 +1,146 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "type": "object",
+  "properties": {
+    "ts": {
+      "type": "integer",
+      "description": "Timestamp in Unix epoch seconds"
+    },
+    "gpu": {
+      "type": "object",
+      "properties": {
+        "name": {
+          "type": "string",
+          "description": "GPU name"
+        },
+        "pci": {
+          "type": "object",
+          "properties": {
+            "gen": {
+              "type": "integer",
+              "description": "PCIe generation"
+            },
+            "lanes": {
+              "type": "integer",
+              "description": "PCIe lanes"
+            }
+          },
+          "required": ["gen", "lanes"]
+        },
+        "temp": {
+          "type": "object",
+          "properties": {
+            "edge_c": {
+              "type": "number",
+              "description": "Edge temperature in Celsius"
+            },
+            "junction_c": {
+              "type": ["number", "null"],
+              "description": "Junction/hotspot temperature in Celsius"
+            }
+          },
+          "required": ["edge_c"]
+        },
+        "util": {
+          "type": "object",
+          "properties": {
+            "gpu_pct": {
+              "type": "number",
+              "description": "GPU utilization percentage"
+            }
+          },
+          "required": ["gpu_pct"]
+        },
+        "vram": {
+          "type": "object",
+          "properties": {
+            "used_mb": {
+              "type": "integer",
+              "description": "Used VRAM in MB"
+            },
+            "total_mb": {
+              "type": "integer",
+              "description": "Total VRAM in MB"
+            }
+          },
+          "required": ["used_mb", "total_mb"]
+        },
+        "clocks": {
+          "type": "object",
+          "properties": {
+            "core_mhz": {
+              "type": "integer",
+              "description": "Core clock in MHz"
+            },
+            "mem_mhz": {
+              "type": "integer",
+              "description": "Memory clock in MHz"
+            }
+          },
+          "required": ["core_mhz", "mem_mhz"]
+        },
+        "power": {
+          "type": "object",
+          "properties": {
+            "watts": {
+              "type": "number",
+              "description": "Power consumption in Watts"
+            }
+          },
+          "required": ["watts"]
+        },
+        "fan": {
+          "type": "object",
+          "properties": {
+            "pct": {
+              "type": "number",
+              "description": "Fan speed percentage"
+            },
+            "rpm": {
+              "type": "integer",
+              "description": "Fan speed in RPM"
+            }
+          },
+          "required": ["pct", "rpm"]
+        },
+        "throttle": {
+          "type": "object",
+          "properties": {
+            "thermal": {
+              "type": "boolean",
+              "description": "Thermal throttling active"
+            },
+            "power": {
+              "type": "boolean",
+              "description": "Power throttling active"
+            }
+          },
+          "required": ["thermal", "power"]
+        },
+        "engines": {
+          "type": "object",
+          "properties": {
+            "gfx_pct": {
+              "type": "number",
+              "description": "GFX engine utilization percentage"
+            },
+            "compute_pct": {
+              "type": "number",
+              "description": "Compute engine utilization percentage"
+            },
+            "dma_pct": {
+              "type": "number",
+              "description": "DMA engine utilization percentage"
+            },
+            "media_pct": {
+              "type": "number",
+              "description": "Media engine utilization percentage"
+            }
+          }
+        }
+      },
+      "required": ["name", "temp", "util", "vram", "clocks", "power", "fan", "throttle"]
+    }
+  },
+  "required": ["ts", "gpu"]
+}