Skip to content

Fine-Tuning¤

Status: Current runtime fine-tuning surface

There is no standalone artifex.fine_tuning package in the current runtime. Artifex currently ships fine-tuning support only through the reinforcement- learning trainers exported from artifex.generative_models.training.

Available Today¤

The retained fine-tuning surface is the RL training layer under artifex.generative_models.training:

  • REINFORCETrainer for policy-gradient updates on typed sequence rollouts
  • PPOTrainer for clipped policy optimization on SequenceRolloutBatch
  • GRPOTrainer for grouped rollout optimization with explicit prompt grouping
  • DPOTrainer for preference optimization on PreferenceBatch

For shared rollout contracts and orchestration guidance, see the RL Training Guide.

Coming Soon¤

These topics remain relevant to the roadmap but are not shipped yet as runtime modules or trainers:

  • LoRA
  • Prefix Tuning
  • Prompt Tuning
  • Distillation
  • Few-Shot Learning
  • Transfer Learning
  • RLHF

See Planned Modules for the current roadmap status of the broader fine-tuning surface.