Introduction: A resource library dedicated to collecting and researching large language model (LLM) post-training methodologies, including papers, code implementations, benchmarks, and community resources. This resource library covers all aspects from basic research to practical applications, including reasoning capabilities of large language models, reinforcement learning, test time expansion methods, etc.
introduction
With the widespread application of large language models (LLMs) in the field of natural language processing, how to improve their reasoning capabilities, decision-making capabilities and alignment through post-training methods has become a research hotspot. Open source projects on GitHub Awesome-LLM-Post-training It brings together papers, code implementations, benchmarks and resources related to post-LLM training to provide a comprehensive reference for researchers and developers.
project overview
The Awesome-LLM-Post-training project was created by the research team of Mohamed bin Zayed Artificial Intelligence University (MBZUAI) and is based on the paper “LLM Post-Training: A Deep Dive into Reasoning Large Language Models” cite turn0search2. The project aims to systematically organize and share the latest research results on post-LLM training methods, covering the following topics:
-
investigation and research: A collection of survey papers on LLM reasoning, decision-making, reinforcement learning, reward learning, strategy optimization, interpretability, multimodal agents, benchmarking, etc.
-
strategy optimization: Contains key papers on policy optimization, such as “Decision Transformer: Reinforcement Learning via Sequence Modeling” and “Offline RL with LLMs as Generalist Memory”.
-
of interpretability: Contains studies exploring the interpretability of LLM, such as “Agents Thinking Fast and Slow: A Talker-Reasoner Architecture.”
-
multimodal agent: Research involving multimodal reasoning, such as “Diving into Self-Evolving Training for Multimodal Reasoning.”
-
Benchmarks and datasets: Provides benchmarks and datasets to evaluate LLM reasoning capabilities, such as “Big-Math: A Large-Scale, High-Quality Math Dataset for Reinforcement Learning in Language Models.”
-
Reasoning and security: discusses safety issues in the LLM reasoning process, such as “Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable”.
How to use this project
To utilize the Awesome-LLM-Post-training project, you can:
-
Visit the GitHub repository: Go to the project homepage Awesome-LLM-Post-training。
-
Browse the README file: Read the project’s README file to understand the content and organization of each part.
-
View related resources: Review the papers, code implementations and benchmarks under the corresponding topics based on your research interests.
-
Contributions and exchanges: If you have relevant resources or experience, you can participate in community contributions by submitting a pull request or discussing it in issues.
conclusion
The Awesome-LLM-Post-training project provides a centralized platform for researchers and developers to access and share the latest research and resources on post-LLM training methods. By using this project, you can gain an in-depth understanding of various post-LLM training methods and improve the model’s reasoning and decision-making capabilities.
Github：https://github.com/mbzuai-oryx/Awesome-LLM-Post-training
Oil tubing: