stado inference plan
Invocation
stado inference plan [OPTIONS] --host <HOST> --image <IMAGE> --model <MODEL> --revision <REVISION> --port <PORT> --max-model-len <MAX_MODEL_LEN> <NAME>Purpose
Inspect the target and persist an immutable, registry-bound plan
Required inputs and options
| Input or option | Contract |
|---|---|
<NAME> | Positional input; required or optional as shown in the invocation. |
--host <HOST> | Command option; required or optional as shown in the invocation. |
--image <IMAGE> | Digest-pinned vLLM image (`repository@sha256:digest`) |
--model <MODEL> | Command option; required or optional as shown in the invocation. |
--revision <REVISION> | Immutable Hugging Face commit SHA |
--gpu-mode <GPU_MODE> | `exclusive` keeps the GPU reserved; `yieldable` pauses inference whenever an eligible GPU job is queued and resumes it afterward Possible values: - exclusive: The GPU stays reserved for inference: no GPU claim is admitted - yieldable: Inference pauses whenever an eligible GPU job is queued and resumes afterward [default: exclusive] |
--port <PORT> | Port the vLLM endpoint listens on |
--max-model-len <MAX_MODEL_LEN> | Longest context the model is served with |
--kv-cache-memory-gb <KV_CACHE_MEMORY_GB> | Fixed vLLM KV-cache allocation in GiB; omit to use the image policy |
--cache-dir <CACHE_DIR> | Persistent host directory for the Hugging Face model cache |
--json | Command option; required or optional as shown in the invocation. |
Output and state effect
This is an inspection/reporting operation. It emits the result described above and does not change managed state unless an explicit option in this page says otherwise. Pass `--json` when the command exposes that machine-readable option.
Refusals
- Before dispatch, Stado refuses a missing required token shown in the invocation, an unknown option, or a value outside a listed value set; argument-usage failures exit with code 2.