Tag

nanochat is a full-stack, minimal codebase for training, fine-tuning, evaluating, and deploying a ChatGPT-like large language model (LLM) from scratch on a single 8xH100 GPU node for under $100.

SWE-agent is a system designed to empower language model (LM) agents to autonomously perform software engineering tasks. It features a custom agent-computer interface (ACI) that enhances the agent's ability to navigate repositories, create and edit code, and execute programs, achieving state-of-the-art results on the SWE-bench and HumanEvalFix benchmarks. [2, 5, 8]

The OpenAI Agents SDK is a lightweight yet powerful framework for building multi-agent workflows. It is provider-agnostic, supporting the OpenAI Responses and Chat Completions APIs, as well as 100+ other LLMs via integrations like LiteLLM.

Agent Lightning is an open-source framework developed by Microsoft Research for optimizing and training AI agents using reinforcement learning (RL) and other techniques, supporting integration with any agent framework with minimal code changes.

This paper introduces ReAct, an approach that integrates reasoning and acting in large language models (LLMs). ReAct enables LLMs to generate both reasoning traces and task-specific actions in an interleaved manner. This synergy allows reasoning to help induce, track, and update action plans, while actions interface with external sources like knowledge bases to gather more information, overcoming issues of hallucination and error propagation in prior methods.

LightRAG is an open-source framework designed for simple and fast Retrieval-Augmented Generation (RAG), integrating knowledge graphs, vector search, and efficient LLM-based processing to enhance question-answering over large document collections.

nanoGPT is the simplest, fastest repository for training/finetuning medium-sized GPTs. It is a rewrite of minGPT that prioritizes practicality over education. Still under active development, but currently the file train.py reproduces GPT-2 (124M) on OpenWebText, running on a single 8XA100 40GB node in about 4 days of training.

KTransformers is a flexible framework for experiencing cutting-edge optimizations in LLM inference and fine-tuning, focusing on CPU-GPU heterogeneous computing. It consists of two core modules: kt-kernel for high-performance inference kernels and kt-sft for fine-tuning. The project supports various hardware and models like DeepSeek series, Kimi-K2, achieving significant resource savings and speedups, such as reducing GPU memory for a 671B model to 70GB and up to 28x acceleration.

Cognee is an open-source tool that provides persistent and dynamic AI memory for agents by combining vector search with graph databases, replacing traditional RAG systems with scalable ECL pipelines.

MiniMind is an open-source GitHub project that enables users to train a 26M-parameter tiny LLM from scratch in just 2 hours with a cost of 3 RMB. It provides native PyTorch implementations for Tokenizer training, pretraining, supervised fine-tuning (SFT), LoRA, DPO, PPO/GRPO reinforcement learning, and MoE architecture with vision multimodal extensions. It includes high-quality open datasets, supports single-GPU training, and is compatible with Transformers, llama.cpp, and other frameworks, ideal for LLM beginners.

Memori is an open-source SQL-native memory engine designed for LLMs, AI agents, and multi-agent systems, providing persistent, queryable memory using standard SQL databases with a single line of code integration.

MineContext is an open-source proactive context-aware AI partner designed to bring clarity and efficiency to your work, study, and creation. It captures and understands your digital world context via screenshots and content comprehension (with future support for multi-modal sources like documents, images, videos, code), and proactively delivers high-quality information such as insights, daily/weekly summaries, to-do lists, and activity records using a context engineering framework.