sf serve

Start, list, extend, and stop Playground inference services

See Playground for the workflow walkthrough.

sf serve start

sf serve start <MODEL> [options]

Argument / Option	Description
MODEL (required)	HF id / absolute path on the shared disk / run:<run_id> (uses that run’s hf_export)
-g, --gpus	GPU count = tensor parallel size (default 1)
--ttl-hours	Idle TTL in hours; defaults to the server-side configuration

sf serve ls / extend / stop

sf serve ls                          # My services: status, endpoint, remaining TTL
sf serve extend <SERVE_ID> --ttl-hours 4
sf serve stop <SERVE_ID>

Services consume real GPU quota; when the TTL expires they stop automatically and release the GPUs. For long sessions use extend, and stop as soon as you’re done.

Services are OpenAI-compatible endpoints — the address from sf serve ls can be configured directly into any OpenAI SDK client.

sf serve

Manage inference services

sf serve <subcommand>

Command	Description
sf serve extend	Extend an inference service’s TTL
sf serve ls	List my inference services
sf serve start	Start an inference service
sf serve stop	Stop an inference service

sf serve extend

Extend an inference service’s TTL

sf serve extend <RUN_ID> [options]

Argument	Description
RUN_ID	required

Option	Description
--hours <float>	How much longer, in hours (default 1.0)

sf serve ls

List my inference services

sf serve ls

No options beyond --help.

sf serve start

Start an inference service

sf serve start <MODEL> [options]

Argument	Description
MODEL	An HF id / an absolute path on shared storage / run:<run_id> (required)

Option	Description
--gpus, -g <int>	How many cards (the tensor-parallel size) (default 1)
--ttl-hours <float>	Idle TTL in hours (default: the server setting)

sf serve stop

Stop an inference service

sf serve stop <RUN_ID>

Argument	Description
RUN_ID	required

No options beyond --help.