Tinygrad 0.11 is out as the newest version of this deep learning framework maintained by Tiny Corp.
Tinygrad 0.11 continues to build out its diverse hardware accelerator support from CPU code to generic OpenCL to supporting different device targets from NVIDIA to AMD to even WebGPU and others like Qualcomm. With the Tinygrad 0.11 release they have added AMD Instinct MI350 accelerator support to their AMD runtime. Their NVIDIA runtime has also added NVIDIA Blackwell GPU support. Additionally, there is now a user-space driver implementation for NVIDIA GPUs too for avoiding the kernel driver.
Tinygrad 0.11 also merged ONNX support for further expanding its software support spectrum.
Tinygrad 0.11 also allows multi-host support via Infiniband, added a Muon optimizer, and various other changes.
Downloads and more details on today’s Tinygrad 0.11 feature release via GitHub.
