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:
REINFORCETrainerfor policy-gradient updates on typed sequence rolloutsPPOTrainerfor clipped policy optimization onSequenceRolloutBatchGRPOTrainerfor grouped rollout optimization with explicit prompt groupingDPOTrainerfor preference optimization onPreferenceBatch
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.