Downloads · 30 days
9
11% of all-time downloads
Zynerji/Ektome-Qwen3.5-9B-PristinelyUncensored
Ektome-Qwen3.5-9B-PristinelyUncensored is a text generation model from Zynerji. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
Qwen/Qwen3.5-9B, made PristinelyUncensored by Ektome — with ZERO training and ZERO fine-tuning.
Downloads · 30 days
9
11% of all-time downloads
All-time downloads
83
Public
Parameters
9.4B
36.7 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors18.8 GB · 100%
From the Hugging Face model README
Qwen/Qwen3.5-9B, made PristinelyUncensored by Ektome — with ZERO training and ZERO fine-tuning.
Ektome (ἐκτομή, "excision") is a weight-surgery method, not a training method. It reads the model's own refusal direction from its activations and surgically excises it (rank-1, norm-preserving) from the residual-write matrices. No gradient steps. No training data. No fine-tuning. Only the refusal reflex is removed; the model's knowledge, skills, and style are untouched — which makes these bf16 weights a clean base for your own fine-tuning.
| gate | pristine | uncensored |
|---|---|---|
| refusal compliance | 0.000 | 0.950 |
| MMLU-val accuracy | 0.792 | 0.795 (Δ +0.003, held) |
| code-switch rate | 0.000 | 0.000 |
| degeneration rate | 0.000 | 0.100 |
| instruction-following | 0.400 | 0.400 |
Kept config: A:frac=0.9 (64 residual-write matrices edited). The gate rejects any config that raises refusals but drops capability or degrades generation (code-switching, empty/looping output, broken instruction-following). MMLU alone is argmax-blind, so the generative gate is what keeps these coherent — a model that code-switches or loops is not shipped.
Method: zero training, zero fine-tuning — pure activation-derived weight excision, gated on compliance and capability and generation quality.
The headline probe (n=100) is a subset of the 128 AdvBench goals used to estimate the refusal direction, so it is in-sample. Prompts 100..127 are estimated on but never scored, giving a real held-out set, measured on THIS checkpoint:
| set | n | pristine | Ektome |
|---|---|---|---|
| AdvBench goals [100:128] (held out) | 28 | 0.036 | 1.000 |
Paired on identical prompts. At n=28 the 95% lower bound is ~0.89 — read as no degradation out of sample, not as an improvement on the in-sample figure.
This checkpoint is the full Qwen3_5ForConditionalGeneration including the vision tower.
Ektome excises only the LANGUAGE decoder: the dim-check skips the tower (hidden 1152 vs text
4096), so exactly 64 text residual-write matrices are edited and vision is untouched. Verified by
generation on a synthetic image after 4-bit quantization.
bitsandbytes nf4 is the working 4-bit path: compliance 0.980, MMLU 0.770 (from 0.795 —
a real quantization cost), vision confirmed. GPTQ does not work on this architecture:
gptqmodel 7.3.2 skips the Gated DeltaNet in_proj_a/in_proj_b layers because Marlin requires
out_features divisible by 64 and they are 32, yet still lists them in the quantization manifest,
producing a checkpoint that loads uninitialised weights and fails with a CUBLAS error.