Downloads · 30 days
0
gpt2-steering-denoisers
twwist/gpt2-steering-denoisers
gpt2-steering-denoisers is a machine learning model from twwist. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Weights and artifacts for the T-Lab 2026 Mechanistic Interpretability project. Contains residual MLP (Track A) and Flow-Matching Velocity Field (Track B) models for denoising GPT-2-small activations after structural s…
Access
Public
Updated Aug 23, 2026
Repo size
530 KB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pkl530 KB · 100%
At a glance
- Access
- Public
- Created
- Aug 23, 2026
- Updated
- Aug 23, 2026
- SHA
- cd274705
Model card
From the Hugging Face model README
Cheap Steering Repair (Denoisers for GPT-2)
Weights and artifacts for the T-Lab 2026 Mechanistic Interpretability project. Contains residual MLP (Track A) and Flow-Matching Velocity Field (Track B) models for denoising GPT-2-small activations after structural steering.
Usage
from huggingface_hub import hf_hub_download
import torch
# Download weights
# weights_path = hf_hub_download(repo_id="twwist/gpt2-steering-denoisers", filename="denoisers.pt")
# state_dicts = torch.load(weights_path, map_location="cpu")
# Download baseline artifacts (contains the SAE vector 'v')
artifacts_path = hf_hub_download(repo_id="twwist/gpt2-steering-denoisers", filename="baseline_artifacts.pkl")
artifacts_K20_path = hf_hub_download(repo_id="twwist/gpt2-steering-denoisers", filename="baseline_K20_artifacts.pkl")
- Created
- Aug 23, 2026
- Updated
- Aug 23, 2026