jhud/mygpt
An easily-trained baby GPT that can stand in for the real thing. Based on Andrej Karpathy's makemore, but set up to mimic a llama-cpp server. This is not production-ready; it's a toy implementation for educational purposes.
25 best points · 4 comments · 1 Hacker News post · 28 GitHub stars · View on GitHub
Hacker News discussions
- Show HN: MyGPT a toy LLM which can be trained on Project Gutenberg and dad jokes (25 points, 4 comments, by disconnection, 2023-09-26)