PyTorch 2.14.0 Released with NVGEMM Kernels Integrated into Inductor and a Preview of the Rewritten NCCL Backend

Written by

in

PyTorch 2.14.0 has been officially released, bringing CUTLASS kernels generated by CuTeDSL into Inductor through NVGEMM, with support for epilogue fusion, scaled GEMM, and NVFP4 GEMM, alongside automatic tuning with Triton and ATen.

Source:PyTorch 正式版本发布 · Read original

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *