Skip to main content
The hud rl command group submits tasks for reinforcement learning training and checks job status.
RL training requires a trainable model. Fork one from Explore at hud.ai/models, then use its model ID here.

​
hud rl run

Submit a set of tasks for RL training with preflight validation. 
hud rl run <source> [OPTIONS]

​
Arguments

​
source
string
Task source: a local file (JSON/JSONL) or a remote taskset name on the platform.

​
Options

​
--model-id, -m
string
Model ID to train. If omitted, an interactive model selector appears.
​
--reasoning-effort
string
default:"medium"
Reasoning effort level: low, medium, or high.
​
--yes, -y
boolean
default:"false"
Auto-accept all prompts (useful for CI).

​
What happens

  1. Model selection — picks the model to train (interactive or via --model-id)
  2. Task loading — loads tasks from a local file or platform taskset
  3. Preflight validation — checks that all referenced environments exist and scenarios match
  4. Submission — sends the training job to the RL service

​
Examples

# Interactive model selection
hud rl run my-taskset

# Specify model directly
hud rl run my-taskset -m mdl_abc123

# From a local file
hud rl run tasks.json -m mdl_abc123

# Skip confirmation (CI mode)
hud rl run my-taskset -m mdl_abc123 -y

# High reasoning effort
hud rl run my-taskset -m mdl_abc123 --reasoning-effort high

​
hud rl status

Check the status of an RL training job. 
hud rl status <model_id> [OPTIONS]

​
Arguments

​
model_id
string
Model ID or job ID to check status for.

​
Options

​
--verbose, -v
boolean
default:"false"
Show full status details in an interactive JSON viewer.

​
Examples

# Check training status
hud rl status mdl_abc123

# Verbose output with full details
hud rl status mdl_abc123 -v

​
Typical Workflow

# 1. Deploy your environment
hud deploy

# 2. Sync tasks to the platform
hud sync tasks my-taskset

# 3. Run evals to generate baseline traces
hud eval my-taskset claude --full

# 4. Submit for RL training
hud rl run my-taskset -m mdl_abc123

# 5. Check progress
hud rl status mdl_abc123

# 6. Evaluate the trained model
hud eval my-taskset -m mdl_abc123 --full

​
See Also