Skill UI

torchforge is Meta's advanced PyTorch-native library designed for agentic Reinforcement Learning (RL). It fundamentally separates core RL algorithms from complex distributed infrastructure concerns. This allows researchers to rapidly prototype and experiment with novel RL methods (like GRPO) using clean, PyTorch-native abstractions, while automatically handling large-scale tasks like distributed training across thousands of GPUs using Monarch and TorchTitan.

PufferLib is a high-performance library designed for advanced reinforcement learning simulations and training. It excels in parallel environment simulation, offering optimized vectorization and native support for multi-agent systems. Utilizing algorithms like PuffeRL (optimized PPO+LSTM), it achieves millions of steps per second, significantly accelerating research and deployment across single-agent and multi-agent environments.

Stable Baselines3 is a robust, PyTorch-based library providing production-ready implementations of core reinforcement learning algorithms (PPO, SAC, DQN, etc.). It facilitates the training of single-agent RL models, supports the creation of custom Gymnasium environments, and handles advanced parallel training via vectorized environments. It is ideal for rapid prototyping and rigorous RL experiments.

This guide provides comprehensive, version-aware documentation for using PufferLib in reinforcement learning pipelines. It details best practices for adapting Gymnasium/PettingZoo environments, managing different PufferLib versions (3.0.0 vs 4.0 source), and executing safe, reproducible training and evaluation routines. Key focus areas include environment contract validation, sandboxing native builds, and maintaining data provenance.

Stable Baselines3 is a PyTorch-based library providing production-ready implementations of core reinforcement learning algorithms, including PPO, SAC, DQN, and TD3. It offers a scikit-learn-like API for easily training RL agents, developing custom Gymnasium environments, and managing complex training workflows. Ideal for standard RL research, rapid prototyping, and advanced model optimization.

This skill provides expert capabilities for training and fine-tuning transformer language models using the TRL (Transformers Reinforcement Learning) library. It supports state-of-the-art post-training methods, including Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), Group Relative Policy Optimization (GRPO), and Reward Model training. Use this skill to align and customize foundation models for advanced tasks via CLI commands.