tinystories-gpt

Tiny GPT (~1.31M params, byte-level BPE vocab 4096) trained on TinyStories for an LLM workshop (chapter 1: run a finished model, then rebuild it).

• Architecture: nanoGPT (Andrej Karpathy, MIT)
• Data: TinyStories (Eldan & Li, 2023)
• Files: ckpt.pt (model_args + weights), tokenizer.json (BPE 4096)

Generates simple children's stories ("Once upon a time, there was a little girl named Lily..."). See the workshop's chapter 1 notebook for loading/usage.