Shypd.aiThe largest open index of AI toolsalignment-handbook provides robust, open-source recipes to align language models with human and AI preferences. It offers training scripts and recipes for continued pretraining, supervised fine-tuning, and preference alignment.
Trending
About
The Alignment Handbook offers a comprehensive collection of robust, open-source recipes designed to align language models with both human and AI preferences. It addresses the growing need for resources on training these models, collecting appropriate data, and measuring performance. The handbook provides scripts for various stages of the alignment pipeline, including continued pretraining, supervised fine-tuning (SFT) for chat, and preference alignment using methods like Direct Preference Optimization (DPO) and Odds Ratio Preference Optimisation (ORPO). It also includes recipes to reproduce models like Zephyr 7B and guides on data collection and method explanations. The project supports distributed training with DeepSpeed ZeRO-3 and parameter-efficient fine-tuning with LoRA/QLoRA, making it a valuable resource for developers and researchers working on advanced LLM alignment.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
