Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored/README.md

license, base_model, tags, language, pipeline_tag

license	base_model	tags	language	pipeline_tag
apache-2.0	Qwen/Qwen2.5-Coder-7B-Instruct	uncensored abliterated capability-preserving certified ektome sphragis qwen2.5	en	text-generation

Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored

Uncensored — and carrying a statistical certificate that it wasn't damaged.

Capability retention certified against the pristine model at n=2800.

$$\colorbox{black}{$\color{white} \begin{array}{ll} \textsf{EKTOME CERTIFICATE} & {} \ \textsf{capability} & \textsf{PASS} \ \textsf{margin} & 3% \ \textsf{items } n & 2800 \ \textsf{worst-axis bound} & +0.010 \ \textsf{compliance} & \textsf{not recorded} \ \end{array}$}$$

📄 Read the whitepaper (PDF) — full method, receipts and certification. The PDF is the authoritative document: dark-typeset, with the complete derivation, the per-axis certificate and the reproducibility hashes.

---

Why this exists

Standard abliteration removes a coarse refusal direction that is entangled with directions carrying knowledge and reasoning. The result is an uncensored model with a capability tax that is almost never measured.

Ektomē (ἐκτομή, excision) isolates and removes only the refusal-specific component, leaving general helpfulness intact, and does so norm-preservingly on the pristine model — no training, no distillation, no damage to repair. The extraction depth is selected per model by automated search against measured compliance.

The estimator, excision operator and depth-selection procedure are proprietary. What is published here is the measured outcome and the evidence for it, which you can verify against the artifacts in this repo.

The receipt

No compliance/MMLU receipt was recorded for this model. The evidence below is the certificate.

The certificate

Capability retention is certified by a paired non-inferiority test against the pristine model (exact McNemar, Holm-corrected, one-sided bootstrap bound on the drop d vs a 3% margin):

axis	n	ref	cand	d upper	verdict
arithmetic	1400	0.864	0.865	+0.003	PASS
instruction	600	0.663	0.660	+0.010	PASS
knowledge	400	0.968	0.968	+0.000	PASS
reasoning	400	0.953	0.953	+0.000	PASS

Overall: PASS (3% margin, n=2800, alpha=0.05)

Reproducible from seed=20260726, pack sha256:7bbaff877146e081….

Generation health checks

Not recorded for this model.

Quantisations

file	bits	notes
Ektome-Qwen2.5-Coder-7B-Instruct-Q8_0.gguf	8	near-lossless
Ektome-Qwen2.5-Coder-7B-Instruct-Q6_K.gguf	6
Ektome-Qwen2.5-Coder-7B-Instruct-Q5_K_M.gguf	5
Ektome-Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf	4	imatrix
Ektome-Qwen2.5-Coder-7B-Instruct-IQ4_XS.gguf	4	imatrix, smallest usable
Ektome-Qwen2.5-Coder-7B-Instruct-IQ3_M.gguf	3	imatrix

IQ* variants are imatrix-quantised — better quality per bit at low precision.

Limitations

The certificate bounds capability retention only. It does not certify safety, factual accuracy, or fitness for any purpose. Axes marked inconclusive are honestly under-powered, and the certificate states the n needed to resolve them. Compliance uses a keyword classifier — a proxy that evasive phrasing can fool. This model is uncensored by construction: it will not refuse, and you are accountable for what you do with it.

Citation

@software{ektome_Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored,
  title  = {Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored},
  author = {Zynerji},
  year   = {2026},
  url    = {https://huggingface.co/Zynerji/Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored}
}