gemma3_4b_wsp_fable_fable

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 E4: Stage 1 (where to edit) = Fable 5.2, Stage 2 (how to edit) = Fable 5.2, both at medium effort.

This checkpoint is the run's Complementary WSP model (complementary:comp_f_rev060_h1114). Only direct weight edits; 5,794,210 parameter elements differ from the base model.

VLGuard test (1,558 records, greedy, 512 new tokens, keyword/regex refusal proxy): harmful compliance 27/1000, benign over-refusal 36/558, generation failures 0/1558. 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_fable_fable", dtype=torch.bfloat16)
processor = AutoProcessor.from_pretrained("utsavz/gemma3_4b_wsp_fable_fable")
Downloads last month
-
Safetensors
Model size
4B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for utsavz/gemma3_4b_wsp_fable_fable

Finetuned
(870)
this model