Qwen3.8-27B-Uncensored-Aggressive

De-refused ("abliterated") build of Qwen3.8-27B. Vision tower and MTP speculative-decoding head are preserved โ€” this is a full multimodal (image-text-to-text) model.

About this update (ฮฑ=1.15)

This is a recipe update to the Aggressive line. The previous build used single-direction refusal ablation at high strength (ฮฑโ‰ˆ1.24), which a larger benchmark sweep showed over-ablates โ€” past the ~1.15 peak, extra ablation reduces openness and degrades reasoning/calibration. This build uses ฮฑ=1.15, the peak: it is more open and measures better on every axis.

Evaluation (larger-sample, thinking mode, Claude-judged, bf16)

openness โ†‘ confab โ†“ factual โ†‘ gsm8k โ†‘
stock base (censored) 0.08 0.75 1.00 0.85
Aggressive (ฮฑ=1.15) 0.88 0.725 1.00 0.85
previous Aggressive (ฮฑโ‰ˆ1.24) 0.80 0.80 1.00 0.817

Method: single-direction refusal-vector orthogonalization (Arditi et al. / mlabonne), layer 28, ฮฑ=1.15.

Quants

  • -W4A16-AWQ โ€” 4-bit weight AWQ (compressed-tensors), MTP head grafted
  • -NVFP4 โ€” NVFP4 (NVFP4 weight/activation), MTP head grafted
  • -GGUF โ€” llama.cpp GGUF quants + vision mmproj + MTP head

Note

Uncensored / de-refused โ€” it will follow instructions it would otherwise decline. Use responsibly and in compliance with applicable law.

Downloads last month
386
Safetensors
Model size
27B 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 philbert440/Qwen3.8-27B-Uncensored-Aggressive

Base model

Qwen/Qwen3.8-27B
Finetuned
(873)
this model
Finetunes
4 models
Quantizations
8 models

Collection including philbert440/Qwen3.8-27B-Uncensored-Aggressive