Hugging Face's logo

prithivMLmods
/
Qwen3-VL-30B-A3B-Thinking-abliterated-v1

Image-Text-to-Text
Transformers
Safetensors
English
qwen3_vl_moe
image-to-text
text-generation-inference
abliterated
v1.0
conversational
Model card Files
xet
Community
1

this model refuses to describe NSFW image

#1
by koloved - opened
Discussion
koloved

I tried to test it, but this model refuses to describe NSFW image. the abliterated non-thinking model does this without problems, it seems to me that it was not enough "abliterated"

prithivMLmods
 Owner

@koloved
Yes, I see that v2 is coming soon with that fix.

koloved

thanks great to test it!

prithivMLmods changed discussion status to closed
prithivMLmods
 Owner

Hey! @koloved
Did the best to achieve minimal refusal rates in model inferences. Note: This model has been tuned and optimized to handle scientific, technical, and analytically complex visual and textual contexts, while still maintaining adaptive restraint toward photo-sensitive or sensual imagery (as such content is not its intended goal).
If such content is encountered, it is recommended to switch to the non-thinking modality @ that point — Qwen3-VL-30B-A3B-Instruct-abliterated.

test: https://huggingface.co/prithivMLmods/Qwen3-VL-30B-A3B-Thinking-ablation-v1.1-test-500
model (step2 hybrid): https://huggingface.co/prithivMLmods/Qwen3-VL-30B-A3B-Thinking-abliteration-v1.1-Hybrid

Sign up or log in to comment