vllm-helper recipes launch
Launch a tested recipe, and point the active profile and Coder at it
Launch a tested recipe, and point the active profile and Coder at it.
To bring the current model back up without switching, use "vllm-helper restart": it drives whichever runner the active profile belongs to.
Usage
vllm-helper recipes launch <id> [flags]
Options
--container-name
| Type | string |
| Default | vllm-node |
container name on each node
--dry-run
| Type | bool |
print what would run and change nothing
--iface
| Type | string |
cluster interface to use instead of the discovered one
--master-port
| Type | int |
| Default | 29501 |
rendezvous port for the ranks
--no-activate
| Type | bool |
launch only; leave the active profile and Coder alone
--nodes
| Type | stringSlice |
| Default | [] |
registry node names in rank order; rank 0 serves the API
--ssh-key
| Type | string |
private key for reaching non-head nodes