Downloads · 30 days
0
OliverSieweke/oracle-encodings-checkpoints
oracle-encodings-checkpoints is a machine learning model from OliverSieweke. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Model checkpoints from the oracle-encodings modular-addition experiments (frozen "oracle" Fourier features injected into the residual stream of a 1-layer transformer grokking (i + j) mod p).
Downloads · 30 days
0
Access
Public
Updated Jun 28, 2026
Repo size
497 MB
Likes
0
Public
Click a slice to open those files.
.pth497 MB · 87%
From the Hugging Face model README
Model checkpoints from the oracle-encodings modular-addition experiments (frozen "oracle" Fourier features injected into the residual stream of a 1-layer transformer grokking (i + j) mod p).
Layout: run_<timestamp>/<experiment>/checkpoints/<label>/ep<NNNNNN>.pth —
one run_<ts>/ tree per runner invocation; the label encodes the sweep axes
(e.g. n2_s0 = 2 injected frequency pairs, seed 0; delay4000_n2_s1,
amp2_n3_s0, rel0.5_n2_s2, ...). Each .pth holds
{model: state_dict, config: Config fields, label, epochs_done} — the frozen
oracle is NOT in the state_dict; rebuild it from the run spec embedded in the
sibling <experiment>/<label>.result.json
(modular_addition.oracle.sweep.build_oracle).