- Models — the models available to dispatch
- On-Demand Capacity — purchasable GPU types with their current
$/GPU-hour, and whether a run starts now or queues
--output json for machine-readable output. For pricing mechanics, see Billing.
The weight cache
Each cluster keeps a cache of model weights. A model listed byprime train models is cached and starts warm. What happens to a model that is not listed depends on where the run goes:
- On-demand runs download the weights from Hugging Face at start-up, then train as normal.
- Runs on a reserved cluster are refused at dispatch, before any GPU is reserved, unless the cluster has the model cached or the run loads it from a volume.
GPU types
To list every GPU type you can dispatch on:--gpu-type, or let the platform pick. Placement favours clusters that already have your model cached, since a cache miss means a full weight download.
Choosing a model
Validate first. Start with the smallest model that exercises your environment end to end — a short run on a small model catches config and reward-function bugs for a fraction of the cost. Keepmax_steps low and watch that the loss or reward moves at all.
Then scale. Once the environment is proven, move to the model you actually care about and raise batch_size and group_size for a more stable reward signal. MoE models with small active parameter counts give strong performance per GPU-hour.
Size the run with the [deployment] block — see Hosted RL for single-node and multi-node examples.
Thinking mode
Some model families support a thinking mode that produces extended chain-of-thought before the final answer. It tends to help on tasks that benefit from multi-step reasoning — math, code, logic — at the cost of longer completions, which means more tokens generated per rollout and more wall-clock per step. Thinking is controlled through the renderer and sampling config; see prime-rl Configuration.Billing
How GPU-hours are metered and where rates come from.
Hosted RL
Config schema and dispatch reference.