Usage
Options
string
Template hash to benchmark. One of
--template_hash or --template_id is required.integer
Template ID to benchmark. Wins over
--template_hash if both are passed. One of --template_hash or --template_id is required.string
Comma-separated GPU names, same format as
vastai search offers (e.g. RTX_4090,RTX_3090). An optional Nx prefix per token requests a multi-GPU configuration (e.g. "2x RTX_4090"). Defaults to RTX_5090,RTX_4090,RTX_3090,RTX_A6000.integer
GPUs per instance for tokens without an
Nx prefix. If omitted, the count is auto-sized from the template’s gpu_total_ram filter so VRAM fits.integer
default:"3600"
Per-GPU safety ceiling in seconds. Each rental is torn down at this point even if no
measured_perf was reported.flag
Skip the cost confirmation prompt.