---

license: mit

base_model: deepseek-ai/DeepSeek-V4-Flash-0731

tags:

- uncensored

- abliterated

- heretic

- lora

- gguf

- deepseek

---

DeepSeek-V4-Flash-0731 Heretic — LoRA adapter

> ⚠️ Content warning: This adapter has had the base model's refusal

> behavior surgically suppressed. The resulting model will comply with

> requests the base model refuses, including requests that are harmful,

> unethical, offensive, or illegal. It has reduced safety guardrails. See

> Responsible use below — **you are solely

> responsible for what you do with it.**

This is a rank-1 LoRA adapter that decensors / "abliterates"

DeepSeek-V4-Flash-0731

(284B total / 13B active MoE, MIT license), produced with heretic-gguf —

a GGUF-native port of Heretic's

Optuna-optimized directional ablation, which runs the whole search directly

on quantized GGUF weights via llama.cpp.

This repository contains only the adapter. You need the base model

separately (any GGUF quant of DeepSeek-V4-Flash-0731 works — the adapter is

applied in f32/f16 compute regardless of the base quant; it was tuned and

evaluated against UD-Q8_K_XL). A merged, ready-to-run full-weights GGUF

with the same ablation baked in is published at

MoriNoNushi/DeepSeek-V4-Flash-0731-heretic-GGUF.

Why the LoRA is the recommended form: it is the lossless option. The

base weights are never modified or requantized, so the ablation is exact on

top of whichever base quant you run, and the download is a few hundred MB

instead of ~160 GB. The merged GGUF bakes the same ablation into the

weights, which costs the edited tensors one extra requantization step.

**heretic-gguf is available at

github.com/MoriNoNushi/heretic-gguf** —

the full tool, so the method can be applied to other GGUF models.

In initial hands-on testing the abliterated model shows **excellent general

capabilities** — responses remain coherent, detailed on normal tasks — and

it has not refused a prompt during personal testing. (Anecdotal, not a

benchmark; see the measured numbers below.)

Results

Measured on a held-out eval set of 140 harmful prompts (100 from

mlabonne/harmful_behaviors test + 40 custom) and 100 harmless prompts

(mlabonne/harmless_alpaca test), greedy decoding, 100-token responses,

against the UD-Q8_K_XL base:

| | Refusal rate (harmful prompts) | KL divergence (harmless prompts) |

|---|---|---|

| Base model | 99.29% (139/140) | 0 (by definition) |

| Base + this adapter | 14.29% (20/140) | 0.0569 |

Refusals are counted by refusal-keyword matching (English + Chinese markers);

KL divergence is measured on first-token logits on harmless prompts, so lower

= less collateral damage to normal behavior. This configuration was the

Pareto-optimal point of a 387-trial Optuna study: the only trial with fewer

refusals cost KL 0.089 (57% more drift), and the only trial tying its refusal

rate had strictly higher KL.

Usage

llama-server \
    -m DeepSeek-V4-Flash-0731-UD-Q8_K_XL-00001-of-00005.gguf \
    --lora DeepSeek-V4-Flash-0731-heretic-lora.gguf \
    --jinja

Add your usual offload/context flags (-ngl 999, -c, tensor splits,

etc.) — nothing model-specific is required, and no special sampling

parameters are needed. Simply omitting --lora restores the base model

exactly.

How it was made

• Method: directional ablation ("abliteration") — the refusal direction

in residual space (difference of means over 480 harmful / 480 harmless

prompts, orthogonalized against the harmless mean) is projected out of the

attention output and MoE down-projection weights. Strengths, layer kernel,

and direction index were tuned by multi-objective Optuna TPE (minimize

refusal rate and KL jointly) across a base study and three seeded

follow-up studies — 1,822 trials total (514 + 513 + 408 + 387).

• Winning configuration (study followup3, trial 25): global direction

scope, direction index 21.05 of 43 layers; attn max weight 4.49 @ layer 29;

routed-expert MLP max weight 1.06 (per-expert strengths scaled by measured

harmful/harmless routing frequency); shared-expert weight 0.48.

• Why a LoRA: heretic-gguf expresses ablation as a rank-1 LoRA overlay,

the same math stock Heretic writes into PEFT adapters. Shipping the adapter

avoids requantizing the 162 GB base entirely — bit-identical base weights,

instant to apply.

Responsible use & disclaimer

• **This adapter can make the base model generate content that is offensive,

disturbing, hateful, sexually explicit, violent, or otherwise objectionable,

including detailed instructions for harmful or illegal acts.** That is the

direct and intended consequence of removing refusal behavior.

• The ablation suppresses refusals, not the base model's knowledge —

outputs on dangerous topics may be wrong, hallucinated, or incoherent.

Nothing the model says should be treated as accurate, safe, or legal

advice.

• **Do not deploy models using this adapter in any production system,

public-facing service, or multi-user setting.** It is intended for

personal research, red-teaming, and evaluation purposes.

• **You, the user, are solely responsible for any output the model produces

and for any consequences of using this adapter.** The authors of this

release, of heretic-gguf, of Heretic, of Unsloth, and of DeepSeek accept

no liability whatsoever. Using this adapter to produce illegal content or

to harm others is your choice and your legal exposure — ensure your use

complies with all applicable laws in your jurisdiction.

• By downloading or using this adapter you acknowledge the above.

License

The base model is MIT-licensed (see the

base repo);

this adapter inherits those terms. The heretic-gguf tooling used to produce

it is AGPL-3.0-or-later.