Latest Advances on System-2 Reasoning

Explore the Multimodal “Aha Moment” on 2B Model

Collect every awesome work about r1!

Official codebase for "Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling".

Doge Family of Small Language Model

Model Context Protocol server for DeepSeek's advanced language models

SOTA RL fine-tuning solution for advanced math reasoning of LLM

Notebooks to demo the use of Azure AI Python SDK / LangChain with DeepSeek R1 reasoning model in Azure AI Foundry.

使用langchain进行任务规划，构建子任务的会话场景资源，通过MCTS任务执行器，来让每个子任务通过在上下文中资源，通过自身反思探索来获取自身对问题的最优答案；这种方式依赖模型的对齐偏好，我们在每种偏好上设计了一个工程框架，来完成自我对不同答案的奖励进行采样策略

Auto-generate fallback and meter display from existing group info in d&b audiotechnik's R1 and ArrayCalc software.

A framework making it effortless to convert any llm model into a reasoning agent like o1 or DeepSeek's r1

Recreating the minimal training methods of DeepSeek-R1 for small langauge models.

斐讯R1音箱控制程序

Latest Advances on (RL based) Multimodal Reasoning in Multimodal Large Language Models

Python Sample App for SMO Systems like Ericsson Intelligent Automation Platform. We aim to be ORAN aligned. Use this to kickstart your own app!

超简单复现Deepseek-R1-Zero和Deepseek-R1，以「24点游戏」为例。通过zero-RL、SFT以及SFT+RL，以激发LLM的自主验证反思能力。

A multi-stage pipeline that enhances Qwen2.5 language models with DeepSeek Reasoner's chain-of-thought capabilities. Implements the DeepSeek-R1 methodology through cold-start SFT, reasoning-oriented RL, rejection sampling, and optional model distillation.

Distributed Reinforcement Learning for LLM Fine-Tuning with multi-GPU utilization

R1 Protocol

LLM public goods game