# Small Stories — third-party notices

Model and tokenizer: Andrej Karpathy, karpathy/tinyllamas, revision `0bd21da7698eaf29a0d7de3992de8a46ef624add`.
The upstream model card declares MIT for this repository. The original model card,
training notes, binary checksums, and MIT notice accompany the pinned files in
`/data/small-stories/0bd21da7698eaf29a0d7de3992de8a46ef624add/`.

JavaScript inference is derived from the MIT-licensed karpathy/llama2.c architecture.
The original copyright and MIT permission notice are retained in engine.js and LICENSE.txt.

Dataset paper (cited, not redistributed): Ronen Eldan and Yuanzhi Li (2023).
TinyStories: How Small Can Language Models Be and Still Speak Coherent English?
https://arxiv.org/abs/2305.07759

Grouped input-embedding ablation is implemented locally; no Captum code is bundled.