|
Download README.md from OX-PIXL/MMDiff_SAEs: direct link, hf CLI and curl.
- Browser
- Download file 1.89 kB
-
https://huggingface.co/OX-PIXL/MMDiff_SAEs/resolve/refs%2Fpr%2F1/README.md
- Command line
-
hf download hf://OX-PIXL/MMDiff_SAEs@refs/pr/1/README.md
-
curl -L -o README.md https://huggingface.co/OX-PIXL/MMDiff_SAEs/resolve/refs%2Fpr%2F1/README.md
1.89 kB
| license: apache-2.0 | |
| pipeline_tag: other | |
| tags: | |
| - sparse-autoencoder | |
| - interpretability | |
| - multimodal | |
| - paligemma2 | |
| - llava | |
| # MMDiff — Multimodal SAE checkpoints | |
| This repository contains the SAE checkpoints used in the paper [Multimodal Model Diffing for Feature Discovery and Control](https://huggingface.co/papers/2608.09928). | |
| Text-only-masked SAEs trained on the residual streams of multimodal LLMs, used by the | |
| **MMDiff** multimodal model-diffing pipeline (code: | |
| [hunarbatra/MMDiff](https://github.com/hunarbatra/MMDiff), project page: | |
| [hunarbatra.com/MMDiff](https://hunarbatra.com/MMDiff/)). | |
| ## Contents | |
| ### `paligemma2/jumprelu_text_only/` — 26 layers | |
| JumpReLU SAEs (width 16,384) trained on `google/paligemma2-3b-mix-448` language-model | |
| residual stream, layers 0–25, with **text-only masking** (image-token positions are | |
| excluded from the training loss). Diffed against the Gemma-Scope base SAEs | |
| (`google/gemma-scope-2b-pt-res`, `width_16k`). | |
| File: `text-only_layer_{L}.pt` with keys `W_enc`, `b_enc`, `W_dec`, `b_dec`, `threshold`. | |
| ### `llava_more/topk_text_only/` — 32 layers | |
| TopK SAEs (k=50, width 32,768) trained on | |
| `aimagelab/LLaVA_MORE-llama_3_1-8B-finetuning` residual stream, layers 0–31, with | |
| text-only masking. Diffed against the LLaMA-Scope base SAEs | |
| (`fnlp/Llama3_1-8B-Base-LXR-8x`, sae-lens release `llama_scope_lxr_8x`). | |
| File: `text-only_layer_{L}.pt` with keys `W_enc (4096×32768)`, `b_enc (32768)`, | |
| `W_dec (32768×4096)`, `b_dec (4096)`. | |
| ## Loading | |
| ```python | |
| import torch | |
| from huggingface_hub import hf_hub_download | |
| p = hf_hub_download( | |
| "OX-PIXL/MMDiff_SAEs", | |
| "paligemma2/jumprelu_text_only/text-only_layer_13.pt", | |
| ) | |
| state = torch.load(p, map_location="cpu") | |
| ``` | |
| See the [MMDiff repo](https://github.com/hunarbatra/MMDiff) for the full pipeline | |
| (training, adapted-feature selection, task-specific discovery, ablation, steering). |