BlueForge Docs

forge deploy

Deploy a compute template.

Command

forge deploy <template> [--gpu <type>] [--count <n>] [--env KEY=VAL] [--cpu <n>] [--memory <n>] [--port <n>] [--timeout <sec>] [--max-cost <usd>] [--wait] [--service|--job]

Arguments

NameTypeRequiredNotes
template
stringyesCompute template name

Flags

FlagTypeDefaultNotes
--gpu
enum—
A10G
| 
A100
| 
H100
| 
RTX4000
--count
int—GPU count
--env
string[]—Environment variables (repeatable: 
--env KEY=val
)
--cpu
int—CPU cores
--memory
int—Memory in GB
--port
int[]—Ports to expose (repeatable)
--timeout
int—Timeout in seconds
--max-cost
float—Max cost in USD (required unless template sets default)
--wait
bool
false
Poll until terminal state
--service
bool—Long-running service
--job
bool—One-off job (default)

Examples

Deploy a CPU job:

forge deploy my-template --cpu 4 --memory 8 --max-cost 5.00

Deploy a GPU service:

forge deploy training-job --gpu A100 --count 1 --service --wait

With environment variables:

forge deploy my-template --env MODEL=llama --env BATCH_SIZE=32

Errors

Exit codeMeaning
0Deploy triggered successfully
1Missing template name
2Auth failure

Pitfalls

  • Compute credentials required — the API key must have compute access.
  • Cost cap — 
    --max-cost
    is required unless the template defines a default.
  • Service vs job — 
    --service
    keeps the deployment running; 
    --job
    (default) runs to completion.

See also

{auto-injected}

Tested against

{auto-injected}