AirLLM 70B inference with single 4GB GPU

Pre-trained image models using ONNX for fast, out-of-the-box inference.

This repository contains the code for PowerGrids, a Modelica library for electro-mechanical modelling of power systems.

"Open Source Models with Hugging Face" course empowers you with the skills to leverage open-source models from the Hugging Face Hub for various tasks in NLP, audio, image, and multimodal domains.

Fine-tune open-source models with Tinker from inside Pi — managed improve loops, data prep, evals, smoke tests, deploy snippets, and checkpoint chat.

Open-source Llm for Apple silicon devices.

The IVI-RA individual patient simulation model

The IVI-NSCLC simulation model

Explore Mistral AI's extensive collection of models. Learn to select, prompt, and integrate Mistral's open-source and commercial models for tasks like classification, coding, and Retrieval Augmented Generation (RAG).

Automate tasks with specialized AI Agents using CrewAI, Langchain and LLMs, a team of agents that can work together to complete tasks using AI prompting techniques from creating tasks to generating keynote speeches.

Resource-efficient LLM distillation: Improving sustainability and reducing computational costs of Large Language Models in financial analytics through knowledge distillation.

Architecting Information for an Open Source Citizenry.

A private, local-first desktop app for chatting with open-source AI models via Ollama — with branching conversations you explore as a visual graph. No cloud, no accounts. macOS · Windows · Linux.

End-to-end demo for deploying and scaling Hugging Face open-source models to production using Foundry Managed Compute in Azure AI Foundry. From Microsoft Build 2026.

🛠️ Manage and sync your coding skills across multiple AI tools with this cross-platform desktop app for streamlined organization and efficiency.

Complete source code for SFT and async GRPO reinforcement learning recipes on Qwen3 models using Microsoft Foundry, Ray, and SLIME — with a multi-turn retail environment and Streamlit dashboard. From Microsoft Build 2026.

🚀 Optimize memory for large language models, enabling 70B models on a 4GB GPU and 405B Llama3.1 on 8GB VRAM without compression techniques.

Enable fast GPU inference of large language models that exceed GPU memory by managing memory dynamically.

AI-powered document revision platform — iterative refinement with structured critique, not ghostwriting

ConvNeXt-CLF-75 is a food image classifier fine-tuned from ConvNeXt_tiny on a curated subset of 75 food categories from the MM-Food-100K dataset. The model is designed to serve as a semantic guide for a downstream segmentation network in a calorie‑estimation pipeline. The classifier outputs class predictions (multi-label).