Efficient GPU kernels for mixed-precision Vision Transformers in Triton - View it on GitHub