Skip to main content
Submits SFT (supervised fine-tuning) or RL (reinforcement learning) training jobs to Synth.

​
Usage

uvx synth-ai train
Finds TOML config files in your directory, prompts you to select one, then submits the training job to Synth. The command polls the job status and shows progress until completion. Inputs:
  • TOML config file (interactive prompt or --config flag)
  • .env file with API keys (from synth-ai setup)
  • For SFT: JSONL dataset file
  • For RL: Hosted task app URL (from modal-serve or deploy)
Outputs:
  • Job submission confirmation
  • Live status updates
  • Job ID and dashboard link

​
Options

  • --config PATH - Specify config file directly
  • --type {auto, sft, rl} - Override auto-detection
  • --dataset PATH - Set dataset file (SFT only)
  • --task-url URL - Set task app URL (RL only)
  • --dry-run - Preview payload without submitting
  • --no-poll - Submit job without waiting for completion
For SFT, the command validates and uploads your JSONL dataset. For RL, it verifies your task app is reachable before submitting.
⌘I