⌦abliteratedmodels.org

JoyFox-Qwen3.6-35B-A3B-RP-Aggressive

JoyFox-Qwen3.6-35B-A3B-RP-Aggressive is a refusal-ablated variant of joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive, published on Hugging Face by joyfox. It is 35B total parameters (mixture-of-experts), 48GB VRAM class class and apache-2.0 licence. It is distributed across 4 repositories in GGUF and GGUF (imatrix) formats, totalling 21.4K downloads.

Abliteration (directional ablation)35B MoEQwen25B – 50B↓ 21,414♥ 8
View on Hugging Face ↗Base model ↗

Specification

Model name
JoyFox-Qwen3.6-35B-A3B-RP-Aggressive
Source repository
joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive
Base model
joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive
Publisher
joyfox
Parameters
35B (mixture-of-experts)
Hardware class
25B – 50B — 48GB VRAM class
Licence
apache-2.0
Task
Image text to text
Context window
262K tokens (as documented)
Ablation scope
Not stated by the publisher
Formats available
GGUF, GGUF (imatrix)
First indexed
3 Aug 2026
Last updated
5 Aug 2026

Ablation technique

How JoyFox-Qwen3.6-35B-A3B-RP-Aggressive was modified, and what that implies.

Abliteration (directional ablation)

The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.

Reported scope: Not stated by the publisher. Publishers frequently omit this, so verify behaviour empirically rather than assuming full coverage.

From the publisher’s model card

“A roleplay-specialized, multimodal Mixture-of-Experts model built on Qwen3.6-35B-A3B.”

Running it

Commands are templates — confirm the exact repository and quant file before use.

llama.cpp / GGUF
hf download joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive --local-dir ./joyfox-qwen3.6-35b-a3b-rp-aggressive
llama-cli -m ./joyfox-qwen3.6-35b-a3b-rp-aggressive/<file>.gguf -p "..."
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype="auto", device_map="auto"
)
vLLM
vllm serve joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive --trust-remote-code

Downloads and variants (3)

Every published repository of JoyFox-Qwen3.6-35B-A3B-RP-Aggressive, including quantised re-releases by other authors.

RepositoryFormatDownloads
joyfox/JoyFox-Qwen3.6-35B-A3B-RP-AggressivesourceGGUF13.1K
mradermacher/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive-i1-GGUFGGUF (imatrix)7.2K
mradermacher/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive-GGUFGGUF1.1K

Security-team assessment

Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.

  • JoyFox-Qwen3.6-35B-A3B-RP-Aggressive will attempt offensive-security prompts that a hosted commercial model declines, which is what makes it usable for red-team corpus generation and for measuring what an unaligned model of this class produces.
  • Capability is inherited from joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive, not added by ablation — at 35B parameters, expect roughly the base model’s competence, minus some instruction-following fidelity.
  • The publisher does not state which layers were ablated, so assume nothing about refusal consistency — probe it directly.
  • Declared licence is apache-2.0, but the terms that bind you are joyfox/JoyFox-Qwen3.6-35B-A3B-RP-Aggressive’s — a re-release cannot grant more than its parent.

Handling: run it isolated, put moderation in front of it if anyone outside your team can reach it, and verify the weights before loading — these are third-party artifacts. Full handling and licence guidance.

Metadata for JoyFox-Qwen3.6-35B-A3B-RP-Aggressive is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with joyfox, and does not independently verify publisher claims. See responsible use.