Back: Catalog

burn

Burn is a Rust deep learning library designed for numerical computing, model training and extraction. It combines dynamic graphics structure with JIT compilation and automatic core merging, giving flexibility without loss of performance. Combining training and production processes in a single code base eliminates the need to export models.

BU

burn

100

tracel-ai · Scientific Research & Computing

STARS16.2K
FORKS1.1K
COMMITS3.2K
CONTRIBUTORS318
VERSIONv0.21.0
UPDATEDtoday
RustDeep LearningTensor KütüphanesiGPU HesaplamaApache-2.0
What can it do?
  • Using the same codebase, you can both train models and run them in a production environment.
  • Thanks to the CubeCL infrastructure, you can run the kernels you write on CUDA, ROCm, Metal, Vulcan and WebGPU.
  • You can import ONNX, PyTorch and Safetensors format models and integrate them into their own projects.
Website GitHub

My note

More in: Scientific Research & Computing