justicedao/legal-ir-autoencoder-checkpoints
Legal IR Autoencoder Checkpoints This dataset stores checkpoint artifacts for the legal text -> formal logic IR autoencoder/Codex optimization loop. Latest checkpoint in this upload: checkpoints/20260630T221836Z/. Contents state/legal-ir-autoencoder-canonical.state.json: canonical feature-level autoencoder warm-start state. reports/: weight review, deprecation manifest, and consensus feature manifests. scripts/review_autoencoder_weight_runs.py: script used to… See the full description on the dataset page: https://huggingface.co/datasets/justicedao/legal-ir-autoencoder-checkpoints.
Legal IR Autoencoder Checkpoints
This dataset stores checkpoint artifacts for the legal text -> formal logic IR autoencoder/Codex optimization loop.
Latest checkpoint in this upload: checkpoints/20260630T221836Z/.
Contents
state/legal-ir-autoencoder-canonical.state.json: canonical feature-level autoencoder warm-start state.reports/: weight review, deprecation manifest, and consensus feature manifests.scripts/review_autoencoder_weight_runs.py: script used to classify runs and build the canonical state.smoke/: summaries/logs from short local smoke tests proving default canonical warm-start loading.manifest.json: checksums, source run IDs, merge weights, and reproducibility metadata.
Canonical State
- Source runs merged:
10 - Reviewed runs:
24 - Deprecated runs:
14 - Merge mode:
weighted_union - Reusable entries:
1205336 - State SHA-256:
7236de26bd3d7f8414ffa04805f1b6e8a8849f9e0103cec6edb4985b911658be
The state is built through ModalAutoencoderTrainingState.load_json(path).generalizable_copy(), so it intentionally excludes sample-specific decoded embeddings, sample family logits, and applied TODO IDs.