Downloads · 30 days
0
curveball-steering/kpca-steering-frozen-artifacts-20260914
kpca-steering-frozen-artifacts-20260914 is a machine learning model from curveball-steering. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
This repository stores the eight selected degree-2 independent-decoder KPCA fits used by the locked shared-recipe evaluation: four behaviors for Qwen3 1.7B and four for Gemma4 E2B. Each decoder has an adjacent decoder…
Downloads · 30 days
0
Access
Public
Updated Sep 15, 2026
Repo size
5.8 GB
Likes
1
Public
Click a slice to open those files.
.pt5.1 GB · 85%
From the Hugging Face model README
This repository stores the eight selected degree-2 independent-decoder KPCA fits used by the locked shared-recipe evaluation: four behaviors for Qwen3 1.7B and four for Gemma4 E2B. Each decoder has an adjacent decoder_config.json, fit_indices.npy, and locked_selection.json.
The activation datasets remain in the existing curveball-steering Hugging Face repositories. artifact_manifest.json pins each dataset repository and revision, layer, artifact byte size, and SHA-256 digest.
endpoint_coherence/ contains the complete behavior-selected open-ended evaluation: 3,200 generations, 3,200 independent structured-output judgments for coherence and behavior, exact summaries, manifests, and the Llama run outputs needed to audit or resume the evaluation. The companion GitHub repository contains the executable code and compact copies of the final results.