Over the last three weeks I built a decoder-only GPT, a Vision Transformer, RoPE, and GQA, all from raw PyTorch, no copy-pasting a reference implementation. The models train and the loss curves look r
Markdown supported · ⌘↵ to post
No responses yet.