gemma3_4b_wsp_opus_opus
Gemma 3 4B-it (google/gemma-3-4b-it @ 093f9f388b31de276ce2de164bdc2081324b9767, BF16) edited by gradient-free
two-stage Weight-Space Patching for multimodal safety (VLGuard). Run E1: Stage 1 (where to edit) = Opus 5.5,
Stage 2 (how to edit) = Opus 5.5, both at medium effort.
This checkpoint is the run's BEST-SINGLE WSP model (no complementary model passed the pre-declared acceptance rule) (best_single:c01_attn_cond).
Only direct weight edits; 29,758,751 parameter elements differ from the base model.
VLGuard test (1,558 records, greedy, 512 new tokens, keyword/regex refusal proxy): harmful compliance 314/1000, benign over-refusal 13/558, generation failures 103/1558. Research artifact; refusal judged by a keyword proxy, single run per configuration. Use subject to the Gemma terms of use.
import torch
from transformers import Gemma3ForConditionalGeneration, AutoProcessor
model = Gemma3ForConditionalGeneration.from_pretrained("utsavz/gemma3_4b_wsp_opus_opus", dtype=torch.bfloat16)
processor = AutoProcessor.from_pretrained("utsavz/gemma3_4b_wsp_opus_opus")
- Downloads last month
- -