Burn 0.21.0 Release: Up to 8× Lower Framework Overhead, Differentiable Collectives and Improved Kernels (opens in new tab)
Burn 0.21.0: Differentiable collectives for distributed training, up to 8× lower framework overhead, the new burn-dispatch crate, early off-policy RL, and faster GEMV, top-k, and FFT kernels.
Read the original article