joennlae/tensorli
Absolute minimalistic implementation of a GPT-like transformer using only numpy (<650 lines).
90 best points · 18 comments · 3 Hacker News posts · 251 GitHub stars · View on GitHub
Hacker News discussions
- Show HN: less than 650 LOC trainable GPT only using NumPy (90 points, 18 comments, by joennlae, 2023-11-17)
- Show HN: Tensorli - GPT-Like Transformers in Pure NumPy (3 points, 0 comments, by joennlae, 2023-11-13)
- Show HN: Tensorli – NumPy Only Transformer Training (<650 lines) (2 points, 0 comments, by joennlae, 2023-11-16)