Projects sharing features with ToRA

:star: Project Page

Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoning

Implementation of the training framework proposed in Self-Rewarding Language Model , from MetaAI

📑 ArXiv Paper ｜ 🤗 Model Weights

This repository contains projects that aims to equip large-scale pretrained language models with better programming and reasoning skills. These projects are presented by Microsoft Research Asia and Microsoft Azure AI.

Code for paper "LEVER: Learning to Verify Language-to-Code Generation with Execution". LEVER is a simple method that improves the code generation ability of large language models trained on code (CodeLMs), by learning to verify and rerank CodeLM-generated programs with their execution results.…

WizardLM is a large language model and instruction-tuning framework designed to execute sophisticated coding, mathematical, and conversational tasks. It functions as an AI system for mathematical reasoning and code generation, as well as a synthetic dataset generator used to train other language models. The project is distinguished by its evolutionary instruction tuning, which uses a method to rewrite simple instructions into complex tasks. This process expands training dataset difficulty and produces a high volume of open-domain tasks across various difficulty levels. The system covers capa

Deductive Beam Search Decoding Deducible Rationale for Chain-of-Thought Reasoning

Note: https://github.com/kyegomez/tree-of-thoughts CANNOT replicate paper results.

siiRL: Shanghai Innovation Institute RL Framework for Advanced LLMs and Multi-Agent Systems

Skywork-Reward-V2 is a series of eight reward models designed for versatility across a wide range of tasks, trained on a mixture of 26 million carefully curated preference pairs. While the Skywork-Reward-V2 series remains based on the Bradley-Terry model, we push the boundaries of training data…

This repository contains the code for the paper "Learning Planning-based Reasoning with Trajectory Collection and Process Rewards Synthesizing" (EMNLP 2024).

This is the official implementation of Graph of Thoughts: Solving Elaborate Problems with Large Language Models. This framework gives you the ability to solve complex problems by modeling them as a Graph of Operations (GoO), which is automatically executed with a Large Language Model (LLM) as…

Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning

put the dataset under data/ Set the hyperparameters in train.slurm and execute bash train.slurm Set the hyperparameters in trainverifier.slurm and execute bash trainverifier.slurm After fine-tuning, specify the model path in mcts.slurm, execute bash mcts.slurm. Note that the provided script will…

📊 Main Results ✨ Getting Started • 📨 Contact • 🎈 Citation • 🌟 Star History

A domain-agnostic framework enabling VLM self-improvement through competitive visual games

SeRL: Self-Play Reinforcement Learning for Large Language Models with Limited Data

AIDE: AI-Driven Exploration in the Space of Code. The machine Learning engineering agent that automates AI R&D.

Official implementation of our Buffer of Thoughts (BoT) framework (NeurIPS 2024 Spotlight). Affiliation: Peking University, UC Berkeley, Stanford University

Open-Instruct is a distributed training and instruction tuning framework for large language models. It functions as a coordinator for supervised fine-tuning, reinforcement learning from human feedback pipelines, and tool-use training, providing specialized roles for dataset curation and model alignment. The project distinguishes itself through a high-performance training architecture that utilizes actor-based distributed coordination and hybrid sharding to manage large GPU clusters. It implements advanced alignment techniques including direct preference optimization, group relative policy opt

The official repository for "Parallel-R1: Towards Parallel Thinking via Reinforcement Learning".