baby-auti-124M-nanoGPT - First attempt at a special-interest hype character
This is a 124M param nanoGPT (Karpathy) trained 100k iters on FineWeb-Edu, then fine-tuned 300 iters at 1e-5 on a small hand-written Q&A dataset (∼500 lines) to create a character called Baby Auti. Character is inspired by strong special interests, loves trains / Thomas / space, answers in an excited, repetitive style. Not a representation of autistic people in general, just a fun character experiment. Facts will be wrong (Manchester is not the capital of England lol), it hallucinates, it's 124M. Base: gpt2 tokenizer, 12 layers, 768 dim.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support