Benchmarks and Triton kernels that progressively fuse GPT-2 attention/MLP work. The repo ships multiple kernel families (A/B/C) plus a plain PyTorch baseline, a small CLI to run block or full-model ...
Minimalist plotting for Python, inspired by Edward Tufte’s principles of data visualization. Maximising the data–ink ratio: remove non-essential lines, marks, and colours. Content-driven spines and ...