Skill UI

Fine-tune LLMs with TRL’s RLHF toolkit, covering SFT, DPO, PPO/GRPO and reward model training so you can align HuggingFace Transformers with human preferences or feedback in post-training workflows.

Provides expert guidance for GRPO training with TRL, covering dataset prep, reward engineering, structured outputs, and multi-objective alignment when preference pairs are scarce.

High-performance RLHF framework that combines Ray, vLLM, and ZeRO-3 to accelerate PPO/GRPO/RLOO/DPO training for 7B-70B+ models on distributed GPU clusters, providing a single-stack workflow for reward models and policy optimization.

Guides post-training of GLM-class LLMs with Megatron-LM and SGLang, focusing on GRPO rollout generation, custom data buffers, and tight RL scaling workflows for research and production use.

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.

Guides training large language models with Volcano Engine's verl library, covering PPO/GRPO workflows, distributed backends, multi-turn rollouts, reward configs, critic-enabled PPO, and production deployments using Ray workers.

A comprehensive skill for fine-tuning and training large language and vision models using advanced techniques like TRL (SFT, DPO, GRPO) or Unsloth. This process runs entirely on managed Hugging Face cloud GPUs, eliminating the need for local GPU setup. It supports modern training workflows, model conversion (GGUF for local deployment), and ensures seamless artifact storage to the Hugging Face Hub.

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.