[ICLR 2025] COAT: Compressing Optimizer States and Activation for Memory-Efficient FP8 Training

Applied AI experiments and examples for PyTorch

A fast communication-overlapping library for tensor/expert parallelism on GPUs.

DeepEP: an efficient expert-parallel communication library

Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

KernelBench: Can LLMs Write GPU Kernels? - Benchmark with Torch -> CUDA problems

Get your documents ready for gen AI

A Easy-to-understand TensorOp Matmul Tutorial

Ongoing research training transformer models at scale

Fully open reproduction of DeepSeek-R1

Material for gpu-mode lectures

Code release for book "Efficient Training in PyTorch"

这是一个从头训练大语言模型的项目，包括预训练、微调和直接偏好优化，模型拥有1B参数，支持中英文。

Agent framework and applications built upon Qwen>=2.0, featuring Function Calling, Code Interpreter, RAG, and Chrome extension.

[NAACL 2024 Outstanding Paper] Source code for the NAACL 2024 paper entitled "R-Tuning: Instructing Large Language Models to Say 'I Don't Know'"

Awesome-LLM-Robustness: a curated list of Uncertainty, Reliability and Robustness in Large Language Models

LLaVA-Mini is a unified large multimodal model (LMM) that can support the understanding of images, high-resolution images, and videos in an efficient manner.

Stretching GPU performance for GEMMs and tensor contractions.

Open-source evaluation toolkit of large multi-modality models (LMMs), support 220+ LMMs, 80+ benchmarks

📰 Must-read papers and blogs on LLM based Long Context Modeling 🔥

About This sample shows how to deploy glm-edge model series using OpenVINO

Only implemented through torch: "bi - mamba2" , "vision- mamba2 -torch". support 1d/2d/3d/nd and support export by jit.script/onnx;

GLM Series Edge Models

Run Generative AI models with simple C++/Python API and using OpenVINO Runtime

A Python package for extending the official PyTorch that can easily obtain performance on Intel platform

LLM implementation one matrix multiplication at a time

Library for modelling performance costs of different Neural Network workloads on NPU devices