Top MODELS GitHub Repositories & Tools (2026)

NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness.

Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

Langflow is a powerful tool for building and deploying AI-powered agents and workflows.

Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks.

🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

The Web framework for perfectionists with deadlines.

[EMNLP2025] LightRAG: Simple and Fast Retrieval-Augmented Generation

Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.

Own your AI. The native macOS harness for AI agents -- any model, persistent memory, autonomous execution, cryptographic identity. Built in Swift. Fully offline. Open source.

The Ruby-native AI framework. Chats, agents, tools, images, audio, and video through one consistent API, in plain Ruby or Rails.

Fast and accurate AI powered file content types detection

DeepInverse: a PyTorch library for solving imaging inverse problems using deep learning

A source-reviewed gallery of JEV-related projects with 50+ GitHub stars — integrations, tools, open models, experiments, and ecosystem resources. Live gallery: beatapi.io/awesome-jev

JEPA-Anything: Learning Predictive Models across Different Worlds

The free AI already on your Mac. CLI tool, OpenAI-compatible server, and interactive chat — all on-device via Apple Intelligence. No API keys, no cloud, no downloads.

Fast, Flexible and Portable Structured Generation

The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more

The set-and-forget LLM engine for Pascal and Volta: PXQ codec + kernels, auto-tuned per card. Ready-to-run PXQ models in MODELS.md; benchmarks vs llama.cpp in the README.

🚀 Efficient implementations for emerging model architectures

Python Toolkit for Causal and Probabilistic Reasoning