RL4LMs
FreeA modular RL library to fine-tune language models to human preferences
About RL4LMs
A modular reinforcement learning library developed by the Allen Institute for AI (AI2) designed to fine-tune language models according to human preferences. It provides a flexible framework for applying various RL algorithms (e.g., PPO, ILQL) to LM training, integrates with Hugging Face Transformers, and supports custom reward models and environments. Aimed at researchers and practitioners working on alignment and RLHF.
Key Features
Alternatives to RL4LMs
CDial-GPT
A Large-scale Chinese Short-Text Conversation Dataset and Chinese pre-training dialog models
ChatLM-mini-Chinese
中文对话0.2B小模型（ChatLM-Chinese-0.2B），开源所有数据集来源、数据清洗、tokenizer训练、模型预训练、SFT指令微调、RLHF优化等流程的全部代码。支持下游任务sft微调，给出三元组信息抽取微调示例。
Chat-Haruhi-Suzumiya
Chat凉宫春日, An open sourced Role-Playing chatbot Cheng Li, Ziang Leng, and others.
DialoGPT
Large-scale pretraining for dialogue
GPT2-Chinese
Chinese version of GPT2 training code, using BERT tokenizer.
gpt-2-simple
Python package to easily retrain OpenAI's GPT-2 text-generating model on new texts