feat: 推理与评测支持多卡 GPU 选择,训练任务实时 GPU 监控

- 新增 GPU 选择归一化 helper,统一前端各形态的选择(gpu_indices/gpus/gpu_id)
- 评测与推理支持同一节点内多卡选择,校验所选 GPU 空闲后再派发
- 新增 /fine-tune/{id}/gpu-status 接口,训练日志页展示实时 GPU 指标
- 算力节点推理加载支持 CUDA_VISIBLE_DEVICES 多卡可见,nvidia-smi 进程级监控
- GPU 占用跟踪细化为按卡记录,覆盖评测任务、推理模型与对比任务

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
wuyongtao
2026-08-19 10:44:56 +08:00
parent c3e96ae61b
commit 9379f93633
13 changed files with 360 additions and 48 deletions

View File

@@ -42,12 +42,23 @@ export interface FineTunePreflightResult {
sync_results?: Array<Record<string, unknown>>
}
export interface FineTuneGpuStatus {
source: string
items: Array<Record<string, unknown>>
selected_gpus: number[]
error?: string
}
/** 训练任务列表 */
export const getFineTuneList = () => get<FineTuneTask[]>('/fine-tune')
/** 训练任务详情 */
export const getFineTune = (id: string | number) => get<FineTuneTask>(`/fine-tune/${id}`)
/** 获取任务所在 Compute 节点的实时 GPU 指标 */
export const getFineTuneGpuStatus = (id: string | number) =>
get<FineTuneGpuStatus>(`/fine-tune/${id}/gpu-status`)
/** 任务名查重 */
export const checkFineTuneName = (name: string) =>
get<{ exists: boolean }>('/fine-tune/check-name', { name })