sf job

查看

sf job ls                     # 作业列表（含提交历史）
sf job status <JOB_ID>        # 状态详情
sf job logs [JOB_ID]          # 跟随日志；省略 ID 跟最近一个
sf job logs <JOB_ID> -n 0     # 回放全量历史再跟随（默认只回放最后 2000 行）
sf job samples <JOB_ID>       # 最近一次验证的多轮对话轨迹
sf job samples <JOB_ID> --vidx 2 -n 10   # 第 3 轮验证，显示 10 条

`logs` 选项	说明
`-n, --tail`	回放最后 N 行历史再跟随；`-n 0` 全量（默认 2000）

`samples` 选项	说明
`--vidx`	验证轮次下标（默认 -1 = 最近一次）
`-n, --limit`	显示样本条数（默认 6）

控制

sf job stop <JOB_ID>          # 停止（运行中 → 终止）
sf job pause <JOB_ID>         # 暂停：保留 checkpoint，可继续
sf job resume <JOB_ID>        # 从最近 checkpoint 自动续训
sf job delete <JOB_ID>        # 删除已结束作业的记录（运行中需先 stop）

pause / resume 是维护友好的暂停语义：管理员维护模式排空集群时作业也走同一条路径，恢复后自动续训不丢进度。

批量

sf job cancel-all             # 停止我所有运行中 / 排队中的作业
sf job clean                  # 清理已结束作业的显示记录
sf job stop-sweep <SWEEP_ID>  # 一键停止一个 sweep 的全部活跃作业

`sf job`

Manage jobs

sf job <subcommand>

命令	说明
`sf job cancel-all`	Stop every running or queued job of mine
`sf job clean`	Clean up the display records of finished jobs
`sf job delete`	Delete one finished job record
`sf job logs`	Follow a job’s logs
`sf job ls`	List jobs (including the submission history)
`sf job pause`	Pause a job, keeping its checkpoint
`sf job resume`	Resume a paused job
`sf job retry`	Retry a failed or stopped job
`sf job samples`	Show one validation round’s traces
`sf job status`	Show a job’s status
`sf job stop`	Stop a job (running -> terminated)
`sf job stop-sweep`	Stop every active job of a sweep

`sf job cancel-all`

Stop every running or queued job of mine

sf job cancel-all [options]

选项	说明
`--yes`, `-y`	Skip the confirmation

`sf job clean`

Clean up the display records of finished jobs

sf job clean

除 --help 外没有其他选项。

`sf job delete`

Delete one finished job record

sf job delete <JOB_ID>

参数	说明
`JOB_ID`	Job id (必填)

除 --help 外没有其他选项。

`sf job logs`

Follow a job’s logs

sf job logs [JOB_ID] [options]

参数	说明
`JOB_ID`	Job id (see sf job ls); omitted = the latest

选项	说明
`--tail`, `-n` `<int>`	Replay only the last N log lines before following (default 2000; -n 0 for everything) (默认 `2000`)

`sf job ls`

List jobs (including the submission history)

sf job ls [options]

选项	说明
`--all`	Show everything (default: the latest 15)
`--exp`	Only this experiment (full name or last segment)
`--limit`, `-n` `<int>`	How many to show (ignored with —all) (默认 `15`)

`sf job pause`

Pause a job, keeping its checkpoint

sf job pause <JOB_ID>

参数	说明
`JOB_ID`	Job id (必填)

除 --help 外没有其他选项。

`sf job resume`

Resume a paused job

sf job resume <JOB_ID>

参数	说明
`JOB_ID`	Job id (必填)

除 --help 外没有其他选项。

`sf job retry`

Retry a failed or stopped job

sf job retry <JOB_ID>

参数	说明
`JOB_ID`	Job id (必填)

除 --help 外没有其他选项。

`sf job samples`

Show one validation round’s traces

sf job samples <JOB_ID> [options]

参数	说明
`JOB_ID`	Job id (see sf job ls) (必填)

选项	说明
`--vidx` `<int>`	Validation round index (default -1 = the latest) (默认 `-1`)
`--limit`, `-n` `<int>`	How many samples to show (默认 `6`)

`sf job status`

Show a job’s status

sf job status <JOB_ID>

参数	说明
`JOB_ID`	Job id (必填)

除 --help 外没有其他选项。

`sf job stop`

Stop a job (running -> terminated)

sf job stop <JOB_ID>

参数	说明
`JOB_ID`	Job id (必填)

除 --help 外没有其他选项。

`sf job stop-sweep`

Stop every active job of a sweep

sf job stop-sweep <SWEEP_ID> [options]

参数	说明
`SWEEP_ID`	The sweep id (printed when sf sweep submits) (必填)

选项	说明
`--yes`, `-y`	Skip the confirmation

CLI 参考

配置

实验

训练

产物与评测

资源

管理

查看

控制

批量

`sf job`

`sf job cancel-all`

`sf job clean`

`sf job delete`

`sf job logs`

`sf job ls`

`sf job pause`

`sf job resume`

`sf job retry`

`sf job samples`

`sf job status`

`sf job stop`

`sf job stop-sweep`

​查看

​控制

​批量

​sf job

​sf job cancel-all

​sf job clean

​sf job delete

​sf job logs

​sf job ls

​sf job pause

​sf job resume

​sf job retry

​sf job samples

​sf job status

​sf job stop

​sf job stop-sweep

查看

控制

批量

`sf job`

`sf job cancel-all`

`sf job clean`

`sf job delete`

`sf job logs`

`sf job ls`

`sf job pause`

`sf job resume`

`sf job retry`

`sf job samples`

`sf job status`

`sf job stop`

`sf job stop-sweep`