vllm.entrypoints.serve.checkpoint.api_router
¶
Functions:
-
checkpoint_prepare–Drain with
modeand release what a checkpoint must not contain. -
checkpoint_restore–Bring back what preparing released and serve. Orchestrators may
checkpoint_prepare(raw_request, mode='wait')
async
¶
Drain with mode and release what a checkpoint must not contain.
Returns once started; poll /checkpoint/status. Repeating is a no-op.
Source code in vllm/entrypoints/serve/checkpoint/api_router.py
checkpoint_restore(raw_request)
async
¶
Bring back what preparing released and serve. Orchestrators may redeliver, so a restore in progress or done returns 200.