trl

FineEnvs — RL Environments 101: building and scaling RL environments in the age of LLMs

Official implementation of GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

LLaMA-TRL: Fine-tuning LLaMA with PPO and LoRA

Notus is a collection of fine-tuned LLMs using SFT, DPO, SFT+DPO, and/or any other RLHF techniques, while always keeping a data-first approach

Agentic RL 中文零基础教程（33 章）：从概念到 GRPO 实战，含 TRL 最小可跑示例；26–33 章附一套可运行的三方判别模型实证工程（encoder vs LLM-LoRA vs 规则基线）。第 25 章讲清 Jev / TypeSafe System One 与 RL 的能力边界 | Chinese Agentic RL tutorial (33 chapters) + a reproducible discriminative-model benchmark

Distill teacher chains-of-thought into a LoRA adapter via a strict boxed-answer format contract + two-phase Train→Nudge (silver-medal NVIDIA Nemotron reasoning recipe, as a tested library).

An implementation of GRPO for Unsloth's VLMs training

基于DPO算法微调语言大模型，简单好上手。

Code repository dedicated to experimenting and research with tiny reasoning language model

RL observability for LLM agents: behavior maps, native trainer metrics, rollout investigations, and Python/SQL analysis.

RL environments with verifiable rewards you can trust. Sandboxed coding and computer-use tasks, verifiers kept apart from the answer, a QA gate against reward hacking, and infra failures kept out of training. Works with verl, TRL, OpenRLHF and Axolotl.

No-code LLM fine-tuning: upload your data → click Train → get a model you can run anywhere. Gradio UI + CLI · SFT/DPO/ORPO/KTO/GRPO/reward models · distillation, vision & embedding models · Unsloth 2-5× · QLoRA · multi-GPU (FSDP/DeepSpeed) · GGUF/FP8 export · vLLM & llama.cpp serving · benchmarks & LLM-judge eval · Heretic Mode

Various training, inference and validation code and results related to Open LLM's that were pretrained (full or partially) on the Dutch language.

simpleR1: A Simple Framework for Training R1-like Models

realize the reinforcement learning training for gpt2 llama bloom and so on llm model

This project demonstrates the process of fine-tuning the Qwen2.5-3B-Instruct model using GRPO (Generalized Reward Policy Optimization) on the GSM8K dataset.

An open-source application that estimates an Open Source Software Technology Readiness Level (OSSTRL) 1–9 from evidence that can be gathered automatically from a GitHub repository.

Training-data memorization auditor for fine-tuned LLMs — Trainer/TRL plugin, canary MIA + regurgitation audit, Apache-2.0

Build task models that replace frontier API calls

Qwen Cup multimodal reasoning competition (20/337 → finals 9/20 → defense 3rd 🏆): teacher-distillation SFT + small-dose GRPO RL on Qwen3.5-0.8B, reward-eval verbatim alignment, full data engineering & ablation chain