gpu

pytorch/pytorch

The Python tensor library behind modern deep learning research — PyTorch combines NumPy-like GPU-accelerated tensors with a tape-based autograd system for building dynamic neural networks. Its imperative, Python-first design lets researchers debug and extend models with familiar tools like NumPy, SciPy, and Cython, making it a core choice for ML experimentation.

Tensors and Dynamic neural networks in Python with strong GPU acceleration

bendlang/bend

Bend 2 pairs dependent-type proofs with CPU/GPU-level speed — Bend 2 is a language meant to let AI write code while humans verify it via mechanical proofs instead of trial-and-error. It claims C-like single-core speed, GPU parallelism with no manual threading, and a proof checker that outperforms assistants like Lean or Agda by orders of magnitude. Aimed at developers who want verifiable AI-generated code.

Bend 2: a fast language that blocks AI mistakes via proof. Install: curl -fsSL https://bend-lang.com/install.sh | sh

gfx-rs/wgpu

The native Rust implementation powering WebGPU in browsers — wgpu implements the WebGPU standard as a native Rust library, running on Vulkan, Metal, D3D12, OpenGL, and WebGL2/WebGPU on wasm. It already powers WebGPU support in Firefox, Servo, and Deno, making it a solid pick for Rust devs building cross-platform graphics or games.

A cross-platform, safe, pure-Rust graphics API.

koharu-rs/koharu

A local-first manga translator built in Rust — Koharu runs detection, OCR, inpainting and LLM translation entirely on-device, keeping manga pages private. It handles raster images, archives and PDFs with page sequencing, and lets users run the pipeline selectively per page or project. Useful for scanlators and hobbyists wanting offline, GPU-accelerated translation with WebGPU.

ML-powered manga translator, written in Rust.

NVIDIA/cuda-rust

NVIDIA opens a Rust path to CUDA kernel programming — NVIDIA's new project brings two GPU programming models to idiomatic Rust: cuda-oxide for traditional SIMT kernels via a custom rustc backend, and cutile-rs for tile-based tensor programming with proc-macros. Rust's ownership rules extend across the GPU launch boundary. Early-stage and experimental, aimed at Rust developers wanting native CUDA access without C++.

NVIDIA's CUDA platform for Rust. Host runtime crates plus Tile (cutile-rs) and SIMT (cuda-oxide) kernel programming models in idiomatic Rust.

TheTumultuousUnicornOfDarkness/CPU-X

CPU-X is a Free software that gathers information on CPU, motherboard and more

SemiAnalysisAI/InferenceX

A standardized benchmark suite for GPU inference stacks — InferenceX from SemiAnalysis bundles end-to-end serving, networking/collective, and kernel-level benchmarks across vLLM, SGLang, and TensorRT-LLM on NVIDIA and AMD GPUs. It targets researchers and MLOps teams comparing inference hardware and software, with results published on a public performance dashboard and new agentic long-context benchmarks like AgentX.

Open Source AI Accelerator Research Platform Standard / 开源推理研究平台

l0ng-ai/tty7

A GPU terminal where shells outlive the window and run agents — tty7 keeps a background server owning shells and panes instead of the window, so sessions survive app quits and reboots without tmux. It detects 28 agent CLIs per pane, lets agents script other agents, and claims ~2x the throughput of Alacritty, Ghostty, or Kitty.

The terminal that outlives its window. Shells survive quitting the app, agents can drive other agents, and remote machines work the same way. Pure Rust, GPU-rendered.

eagle-hq/EAGLE-4

Official Inference Code of EAGLE-4

sorrycc/Mayoi

A small, fast code editor written in Go, with a native GPU-drawn UI (no webview, no Electron)

ymcrcat/rgpu

Offload PyTorch tensors to a remote NVIDIA GPU, no CUDA on client — rGPU lets a PyTorch script run on a Mac or any CUDA-less laptop while tensor ops execute on a remote NVIDIA machine over TCP. It offers a simple `rgpu` device for PyTorch code plus a CUDA shim covering existing Linux CUDA binaries, including libcuda, cuBLAS and cuDNN.

Keep Python on your laptop. Run PyTorch operations and hold tensors on a remote GPU, including from a Mac with no CUDA installation.