metadata
license: mit
pretty_name: Token Learning Spectrum Examples
tags:
- scaling-laws
- language-modeling
- token-loss
- interpretability
Token Learning Spectrum Examples
This dataset hosts public losses.npz matrices for reproducing the figures in
the Token Learning Spectrum code release.
Each losses.npz follows the schema documented in the GitHub repository:
axis_values: float array [K]
loss_matrix: float array [N, K]
axis_name: string array [1]
sample_id, token_pos, and metadata arrays: optional arrays [N]
Files are listed in manifest.yaml with shapes, byte sizes, and SHA256
checksums. Download them with:
python tools/download_examples.py --repo-id applewpj/token-learning-spectrum-examples --all
The T/D/M-axis loss matrices are sanitized public analysis inputs. They do not include private model architectures, weights, tokenizers, raw validation text, training data composition, experiment registries, or checkpoint paths.
The synthetic Mano matrix is generated from the public synthetic arithmetic pipeline adapted from PhysicsLM4.