Open source. Open possibilities.

Official code repository for 'Build a Large Language Model (From Scratch)': step-by-step PyTorch implementation of GPT-like models, covering pretraining and finetuning with notebooks and exercises.

PowerInfer is a local LLM inference engine that uses activation locality to keep frequently used “hot” neurons on the GPU while computing input-dependent “cold” neurons on the CPU. It supports hybrid or CPU-only inference, serving, batching, perplexity evaluation, INT4 quantization, and PowerInfer GGUF models.

The Chinese-LLaMA-Alpaca project enhances the original LLaMA model for Chinese NLP by expanding its vocabulary, conducting secondary pre-training with Chinese corpora, and fine-tuning with instruction data. It offers instruction-tuned models, LoRA weights, training scripts, and quantization/deployment tools for local CPU/GPU execution, supporting various ecosystem tools.

A text-based 5e-compatible RPG where an AI Dungeon Master uses an MCP/SQLite game engine for real dice rolls, persistent stats, spell slots, and combat.

SupaNexus is an open AI gateway that unifies access to multiple LLM providers via one API, simplifying development with OpenAI-compatible interfaces and centralized model management.

LightRAG is a lightweight graph-based retrieval-augmented generation framework from HKUDS, offering dual-level retrieval, knowledge-graph indexing, multimodal parsing, multiple storage backends and an API server with WebUI.

GPT Academic is an open-source project providing practical interaction interfaces for GPT/GLM and other large language models, focusing on paper reading, polishing, and writing. It supports modular plugin systems, custom shortcuts, parallel queries across multiple models, and PDF/LaTeX translation and summarization.

An all-in-one AI framework for semantic search, LLM orchestration, and language model workflows.

A practical, from-scratch guide to training LLMs using PyTorch, covering data preparation, building a Transformer, pretraining, and post-training with SFT, DPO, PPO, and GRPO.

Haystack is an open-source AI orchestration framework for building production-ready LLM applications, RAG systems, and autonomous agents in Python.

LlamaFactory is a Python framework for efficient fine-tuning of 100+ large language and multimodal models, offering zero-code CLI, a Gradio Web UI, LoRA/QLoRA, and multiple training algorithms.

Deep Lake is a serverless multimodal database for AI that combines vector search with storage for embeddings, images, video, audio, and documents. It streams data directly from cloud storage to PyTorch/TensorFlow, supports versioning, and integrates with LangChain, LlamaIndex, and Weights & Biases.

A free educational course on Large Language Models featuring roadmaps and Google Colab notebooks. Covers LLM fundamentals, architecture, pre-training, fine-tuning, quantization, and building LLM-based applications.

Ray is a unified framework for scaling AI and Python applications from a single laptop to a distributed cluster.

Flowise is a visual tool for building AI agents, allowing users to design and deploy agentic workflows through a graphical interface.

Langflow is a visual platform for building, deploying, and orchestrating AI-powered agents and workflows. It offers a drag-and-drop interface, interactive playground, multi-agent support, and can deploy as API or MCP server. Supports all major LLMs and vector databases.

LangExtract is a Python library from Google that uses LLMs to extract structured information from unstructured text, grounding every extraction to its exact source location and generating interactive HTML visualizations.

DeepTutor is an open-source lifelong personalized tutoring platform from HKUDS, combining agentic chat, knowledge bases with multiple RAG engines, guided learning, book generation, quizzes and multi-channel IM partners.