Downloads · 30 days
0
espalier-benchmark/espalier-mib-submission
espalier-mib-submission is a machine learning model from espalier-benchmark. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
This directory contains only the frozen Espalier / Energy + graph circuits for the MIB Circuit Localization edge track.
Downloads · 30 days
0
Access
Public
Updated Sep 3, 2026
Repo size
—
Likes
0
Public
Click a slice to open those files.
.json179 MB · 100%
From the Hugging Face model README
This directory contains only the frozen Espalier / Energy + graph circuits for the MIB Circuit Localization edge track.
ioi_gpt2ioi_qwen2.5mcqa_qwen2.5Each directory contains exactly nine JSON circuits, ordered by the numeric prefixes 01–09: 0.1%, 0.2%, 0.5%, 1%, 2%, 5%, 10%, 20%, and 50% of the canonical real-edge universe.
Source commit: 5ca140221f865d899a07ca5451d8b4433a376b8f.
Every staged membership is derived from the already-frozen evaluation_active circuit state used by the public validation evaluator. GPT-2 files are byte-identical copies of their frozen Graph JSON sources. Qwen frozen artifacts store exact active edge-ID lists; those are converted into current MIB Graph JSON while preserving the membership set exactly. See SUBMISSION_MANIFEST.json for every source and SHA-256.
No private-test information was used. This staging package was produced without rerunning or modifying model evaluation, attribution, circuit discovery, score arrays, or frozen circuit artifacts.