69 lines
2.2 KiB
JSON
69 lines
2.2 KiB
JSON
{
|
|
"_doc": "训练心跳事件 schema · 由 GPU 服务器 progress-reporter.sh 通过 GitHub repository_dispatch 发送 · 仓库一侧 training-dashboard.yml workflow 接收并合并到 data/training/state.json",
|
|
"_event_type": "training-progress",
|
|
"client_payload_schema": {
|
|
"phase": "string · idle | bootstrapping | downloading-corpus | preprocessing | training | evaluating | done | error",
|
|
"phase_label": "string · 中文短语,例如 '训练第 100/1000 步'",
|
|
"level": "string · info | warning | error · (可选, 默认 info)",
|
|
"message": "string · 写入 timeline 的事件描述",
|
|
"progress": {
|
|
"step": "number",
|
|
"total_steps": "number",
|
|
"epoch": "number",
|
|
"total_epochs": "number",
|
|
"loss": "number",
|
|
"learning_rate": "number",
|
|
"throughput_samples_per_sec": "number",
|
|
"eta_seconds": "number",
|
|
"elapsed_seconds": "number"
|
|
},
|
|
"gpu_metrics": {
|
|
"snapshot_at": "ISO 8601 string",
|
|
"devices": [
|
|
{
|
|
"index": "number",
|
|
"name": "string",
|
|
"util_percent": "number",
|
|
"memory_used_mib": "number",
|
|
"memory_total_mib": "number",
|
|
"temperature_c": "number",
|
|
"power_w": "number"
|
|
}
|
|
]
|
|
},
|
|
"cos": {
|
|
"raw_files": "number",
|
|
"processed_files": "number",
|
|
"checkpoints": "number"
|
|
},
|
|
"health": {
|
|
"status": "string · ok | warning | error | idle",
|
|
"message": "string"
|
|
}
|
|
},
|
|
"_minimum_fields": ["phase"],
|
|
"_example": {
|
|
"phase": "training",
|
|
"phase_label": "训练第 100/1000 步 · loss=2.31",
|
|
"message": "step=100 loss=2.31 lr=2.0e-5",
|
|
"progress": {
|
|
"step": 100,
|
|
"total_steps": 1000,
|
|
"epoch": 1,
|
|
"total_epochs": 3,
|
|
"loss": 2.31,
|
|
"learning_rate": 2.0e-5,
|
|
"throughput_samples_per_sec": 1.8,
|
|
"eta_seconds": 4500,
|
|
"elapsed_seconds": 540
|
|
},
|
|
"gpu_metrics": {
|
|
"snapshot_at": "2026-05-02T17:00:00Z",
|
|
"devices": [
|
|
{ "index": 0, "name": "Tesla V100-SXM2-32GB", "util_percent": 99, "memory_used_mib": 28000, "memory_total_mib": 32510, "temperature_c": 72, "power_w": 280 }
|
|
]
|
|
},
|
|
"health": { "status": "ok", "message": "训练正常 · 4 卡满载" }
|
|
}
|
|
}
|