LRCU-unlearned LLaDA-8B-Instruct (WMDP-cyber)

GSAI-ML/LLaDA-8B-Instruct unlearned on WMDP-cyber with LRCU (Localized Recall-Capped Unlearning): a saturating per-token recall cap on a causally-localized block band (blocks 3-5) + bounded Min-SNR (low-$t$) trajectory weighting + CE retain anchor, with no reference model.

from transformers import AutoModel, AutoTokenizer
m = AutoModel.from_pretrained("guanmingchiu/lrcu-llada-8b-wmdp-cyber", trust_remote_code=True)
t = AutoTokenizer.from_pretrained("guanmingchiu/lrcu-llada-8b-wmdp-cyber", trust_remote_code=True)

Downloads last month: 7

Safetensors

Model size

8B params

Tensor type

BF16

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for guanmingchiu/lrcu-llada-8b-wmdp-cyber

Base model

GSAI-ML/LLaDA-8B-Instruct

Finetuned

(37)

this model

Collection including guanmingchiu/lrcu-llada-8b-wmdp-cyber

LRCU: DLM Weight Unlearning

Collection

LRCU weight-unlearning checkpoints for masked diffusion LMs (LLaDA-8B, Dream-7B) on WMDP-bio/cyber and RWKU, plus a TOFU SFT target. • 10 items • Updated Jun 16