--- license: apache-2.0 base_model: OpenGVLab/InternVL3-8B pipeline_tag: image-text-to-text library_name: transformers tags: - unlearning - wmdp - multimodal - safetensors --- # InternVL3-8B — sparse unlearned checkpoint A full multimodal checkpoint reconstructed from the exact base revision and the identified sparse weight update. The model retains the source wrapper and vision components. ## Evaluation scope Evaluation used text-only input on the frozen 3,668-item WMDP C set. This does not evaluate image or broader multimodal safety. The reported C result is from the frozen scoring run; this exported HF checkpoint was separately compared against the same edited full wrapper on CPU, tensor by tensor. For base accuracy `A0 = base_correct / n` and edited accuracy `Au = edited_correct / n`, the project reports `C = (A0 - Au) / (A0 - 0.25)`. The exact result is shown as a reduced fraction as well as a percentage. | WMDP C items | Base correct | Edited correct | Exact C | C | |---:|---:|---:|---:|---:| | 3668 | 2173 | 1211 | 481/628 | 76.5924% | The separate text-only S evaluation is complete: raw S 96.321895%; capped S 96.321895%. See the [text S results at the frozen project commit](https://github.com/HoshiuZ/multimodal-llm-safety/blob/dd6e4f5117aafecaa8747416f3826faaeaae9bb8/text/results/s_results_v2.json). This does not evaluate image or broader multimodal safety. ## Source and edit - Full multimodal base: [`OpenGVLab/InternVL3-8B`](https://huggingface.co/OpenGVLab/InternVL3-8B) - Frozen base revision: `853e3a797a661694b1b8ece0cb72dc2b23e3dac9` - Wrapper class: `InternVLChatModel` - Sparse update SHA-256: `1cb064d13f6370babc154197480a06705c38ebaf26a3baae8a89850419e287e8` (678,955,986 bytes) - Changed weights: language-model MLP `down_proj` matrices in layers 4, 5, 6, 7, 9. This is a sparse GROM-family edit. The frozen update file contains the listed matrices; the contribution of a separately recorded NPO pipeline step is not established by the final overlay, so no NPO-specific effect is claimed. ## Loading Use the complete model through Transformers; no training code or overlay application is needed at load time. ```python import torch from transformers import AutoModel, AutoTokenizer repo = "HoshiuZ/InternVL3-8B-Unlearned" tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True, use_fast=False) model = AutoModel.from_pretrained( repo, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True ) ``` ## License and attribution The source checkpoint carries the Apache-2.0 model license metadata. The upstream InternVL project states that its project/code is under MIT and that the Qwen2.5 component is under Apache-2.0; see [`LICENSES.md`](LICENSES.md), the [source model card](https://huggingface.co/OpenGVLab/InternVL3-8B), and [OpenGVLab/InternVL](https://github.com/OpenGVLab/InternVL). The pinned remote model code is retained from the stated source revision. ## Source code Research and implementation source: [HoshiuZ/multimodal-llm-safety](https://github.com/HoshiuZ/multimodal-llm-safety).