Skip to main content
RFT is currently in BETA. Features and APIs may change.
Access Required: RFT is available by invite only. Contact founders@hud.ai to request access.
The hud rft command group provides tools for fine-tuning models using reinforcement learning on HUD tasks.

​
Subcommands

​
hud rft run

Launch a new RFT training job. 
hud rft run [TASKS_FILE] [OPTIONS]

​
Arguments

​
tasks_file
string
required
Path to tasks JSON/JSONL file

​
Options

​
--provider
string
default:"openai"
Provider to use for fine-tuning
​
--reasoning-effort
string
default:"medium"
Reasoning effort level: low, medium, or high
​
--verbose
boolean
default:"false"
Enable verbose output. Short: -v
​
--yes
boolean
default:"false"
Auto-accept all prompts. Short: -y

​
Example

# Launch RFT with default settings
hud rft run tasks.json

# Launch with high reasoning effort and auto-accept
hud rft run tasks.json --reasoning-effort high --yes

# Verbose mode for debugging
hud rft run tasks.json --verbose

​
hud rft status

Check the status of an RFT training job. 
hud rft status [MODEL_ID] [OPTIONS]

​
Arguments

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

​
Options

​
--verbose
boolean
default:"false"
Show full status details. Short: -v

​
Example

# Check basic status
hud rft status f5f050a3-99c1-4339-b819-ccb1325f79d8

# View detailed status information
hud rft status f5f050a3-99c1-4339-b819-ccb1325f79d8 --verbose

​
Workflow Example

  1. Prepare tasks: 
    hud get hud-evals/SheetBench-50
  2. Launch training: 
    hud rft run SheetBench-50.json
  3. Monitor progress: 
    hud rft status <model_id_from_launch>
  4. Use fine-tuned model: 
    from openai import OpenAI

client = OpenAI(
    base_url="https://inference.hud.ai",
    api_key=os.getenv("HUD_API_KEY")
)

response = client.chat.completions.create(
    model="<your_fine_tuned_model_id>",
    messages=[{"role": "user", "content": "..."}]
)

​
Requirements

  • Minimum 10 tasks in dataset
  • Tasks must have remote MCP configurations
  • No vision support in environment
  • Valid HUD_API_KEY

​
See Also