L

Llama 3.1 Tulu 3.1 8B

Developed by allenai
Tülu 3 is a leading family of instruction-following models, offering fully open-source data, code, and training methodologies as a comprehensive guide to modern technology. Version 3.1 features improvements in the reinforcement learning phase, delivering enhanced overall performance.
Downloads 3,643
Release Time : 2/7/2025

Model Overview

An 8B-parameter instruction-following model based on the Llama 3.1 architecture, designed for diverse tasks (such as mathematics, GSM8K, and IFEval) with excellent performance.

Model Features

Reinforcement Learning Optimization
Version 3.1 switches from PPO to GRPO (without a reward model) and adjusts hyperparameters, achieving comprehensive performance improvements.
Diverse Task Performance
Delivers excellent performance on diverse tasks such as mathematics, GSM8K, and IFEval.
Fully Open-Source
Provides fully open-source data, code, and training methodologies.

Model Capabilities

Text generation
Mathematical reasoning
Code generation
Instruction following

Use Cases

Education
Math problem solving
Solving math problems like GSM8K
Achieves 90.0% accuracy on GSM8K
Programming
Code generation
Generating Python code
Achieves 84.8% pass@10 on HumanEval
Q&A Systems
Knowledge Q&A
Answering various knowledge-based questions
Achieves 69.5% accuracy on MMLU 5-shot
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase