Skip to main content

Roadmap

We’re continuously expanding AITraining’s capabilities. Here’s what’s currently supported and what’s coming next.

Currently Supported

LLM Training

  • SFT - Supervised Fine-Tuning for instruction following
  • DPO - Direct Preference Optimization
  • ORPO - Odds Ratio Preference Optimization
  • PPO - Proximal Policy Optimization (RL)
  • Reward Modeling - Train reward models for RLHF
  • Knowledge Distillation - Transfer knowledge from larger models

Text Tasks

  • Text Classification - Sentiment, spam detection, categorization
  • Token Classification - NER, POS tagging, entity extraction
  • Sequence-to-Sequence - Translation, summarization
  • Extractive QA - Answer questions from context
  • Sentence Transformers - Semantic similarity embeddings

Vision Tasks

  • Image Classification - Categorize images into labels
  • Image Regression - Predict continuous values from images
  • Object Detection - Locate and identify objects in images
  • Vision-Language Models - Multimodal image+text tasks

Tabular Data

  • XGBoost - Gradient boosting
  • LightGBM - Fast gradient boosting
  • Random Forest - Ensemble decision trees
  • CatBoost - Categorical feature handling
  • ExtraTrees - Extremely randomized trees

Reinforcement Learning

  • PPO Trainer - Proximal Policy Optimization for LLMs
  • DPO Trainer - Direct Preference Optimization
  • Reward Models - Standard, pairwise, and multi-objective
  • RL Environments - Text generation, math problems, code generation
  • Async Forward-Backward Pipeline - Efficient training pipeline

Planned Training Tasks

Vision Tasks (Planned)

Time Series & Forecasting (Planned)

Additional ML Algorithms (Planned)

Specialized LLM Training (Planned)

Audio & Speech (Planned)

Multimodal (Planned)

Specialized Domains (Planned)

Planned Features

Training Enhancements

  • Ray Tune integration for distributed sweeps
  • Curriculum learning support
  • Continual learning / catastrophic forgetting prevention
  • Mixture of Experts (MoE) fine-tuning
  • Speculative decoding training

Infrastructure

  • Full TUI (Terminal User Interface) wizard
  • Web-based training UI
  • Kubernetes deployment templates
  • AWS/GCP/Azure marketplace images

Evaluation

  • Automated red-teaming
  • Bias and fairness benchmarks
  • Domain-specific evaluation suites
  • Human preference collection interface

Vote for Features

Want to influence our priorities? Let us know what matters most to you:

GitHub Discussions

Vote on feature requests and propose new ideas

Discord Community

Join the discussion and share your use cases

Contributing

Interested in helping build these features? We welcome contributions:
  • Core Development: Python, PyTorch, Transformers
  • Documentation: Help us document new features
  • Testing: Test new trainers and report issues
  • Examples: Share your training recipes
See our GitHub repository for contribution guidelines.

Release Notes

For current features and recent updates, see the Changelog.