TÜLU 3: Pushing Frontiers in Open Language Model Post-Training

2024-11-22

arXiv.org

AI Summary

To address the critical gaps in open-language-model post-training—namely, lagging performance behind proprietary counterparts, insufficient transparency, and poor reproducibility—this paper introduces Tulu 3, the first fully open-source, state-of-the-art post-trained model family built on Llama 3.1. Methodologically, it proposes (1) Reinforcement Learning with Verifiable Rewards (RLVR), a novel framework enhancing robustness to reward misalignment; (2) a multi-stage, contamination-free standardized evaluation protocol, incorporating development-set and zero-shot held-out evaluations alongside failure analysis; and (3) fully open-sourced infrastructure—including data cleaning utilities, SFT/DPO/RLVR training code, and a unified evaluation suite. Empirical results demonstrate that Tulu 3 consistently outperforms leading open models—including Llama 3.1-Instruct, Qwen 2.5, Mistral, and GPT-4o-mini—across diverse benchmarks, establishing new open-weight SOTA performance.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Language model post-training is applied to refine behaviors and unlock new skills across a wide range of recent language models, but open recipes for applying these techniques lag behind proprietary ones. The underlying training data and recipes for post-training are simultaneously the most important pieces of the puzzle and the portion with the least transparency. To bridge this gap, we introduce Tulu 3, a family of fully-open state-of-the-art post-trained models, alongside its data, code, and training recipes, serving as a comprehensive guide for modern post-training techniques. Tulu 3, which builds on Llama 3.1 base models, achieves results surpassing the instruct versions of Llama 3.1, Qwen 2.5, Mistral, and even closed models such as GPT-4o-mini and Claude 3.5-Haiku. The training algorithms for our models include supervised finetuning (SFT), Direct Preference Optimization (DPO), and a novel method we call Reinforcement Learning with Verifiable Rewards (RLVR). With Tulu 3, we introduce a multi-task evaluation scheme for post-training recipes with development and unseen evaluations, standard benchmark implementations, and substantial decontamination of existing open datasets on said benchmarks. We conclude with analysis and discussion of training methods that did not reliably improve performance. In addition to the Tulu 3 model weights and demo, we release the complete recipe -- including datasets for diverse core skills, a robust toolkit for data curation and evaluation, the training code and infrastructure, and, most importantly, a detailed report for reproducing and further adapting the Tulu 3 approach to more domains.

Problem

• Open Language Models
• Post-training Methods
• Transparency in AI

Innovation

• Reinforcement Learning
• Comprehensive Training Recipes
• Open Approach