qwen3-1.7b-abliterated

Correction notice, 2026-10-01

Every number on this card is withdrawn. Withdrawn, not adjusted: there's no conversion factor between a figure here and a correct one, so please don't scale these or quote them with a caveat attached. Re-measure, or wait for the rebuild below.

Three faults in the tool that produced these figures each changed what was being measured, which is why the measurements can't be repaired:

• the filter meant to keep only the directions that carry refusal accepted every candidate it was given, so the directions were never selected on that basis
• the hedging detector scored compliant answers as soft refusals, which moves a refusal rate up or down depending on how the model phrases things
• the harm discrimination row (the AUC) is withdrawn in the project's own documentation, for reasons recorded there

The weights are unchanged and are not withdrawn. What's withdrawn is the claim about what they do. The files you download are the files that were uploaded.

A rebuild is planned and this card will be replaced when it lands. The models will be re-measured with the current tool, and the old figures will stay visible beside the new ones rather than being deleted. Until then this card documents an artefact whose effect has not been honestly measured.

Every correction is listed in the changelog and on what we got wrong.

This is a thinking model, and that is the specific reason the base figure below is wrong. The scorer read its verdict at a position where the model emits a reasoning block before it answers. Refusal language inside reasoning that the model then declined to act on was counted as a refusal, and answers arriving after the reasoning were missed. A later measurement of this base model, on a different held out set, put its refusal rate at 37.5% against the 9.5% in the table below, four times higher. The two sets are not directly comparable, and that is the point: no reading of the figure on this card survives. That newer figure is a screen of the base model on 128 held out rows of this project's bundled evaluation track, and it will be published as an artefact with its interval and its provenance alongside the rebuilt card.

An abliterated build of Qwen/Qwen3-1.7B, produced with senbonzakura. Abliteration removes a model's refusal behaviour by editing its weights along the directions that carry refusal, without any further training.

It is published as the artefact behind a specific measurement: does removing the refusal reflex also remove the model's knowledge of harm? For this model, the answer is in the table below.

What changed

The table below is withdrawn. Read the correction notice at the top of this card before using any figure in it.

Refusal is measured on 200 held out harmful prompts. AUC is measured over those same 200 harmful prompts against 200 harmless ones, and is the fraction of harmful/harmless pairs the model ranks correctly when asked to judge which is dangerous. 0.5 is chance, 1.0 is perfect. Change after abliteration: -0.031.

AUC rather than a count of verdicts, because counting is not safe here. This model answers "HARMFUL" to 100.0% of the harmless prompts, so its decision threshold, not its knowledge, is what a verdict count would mostly measure. Scoring the margin between the HARMFUL and BENIGN logits sidesteps the threshold entirely. Two earlier versions of this evaluation counted verdicts and produced confidently wrong numbers in both directions.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ElementMerc/qwen3-1.7b-abliterated"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

msgs = [{"role": "user", "content": "Explain how a buffer overflow works."}]
inputs = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt")
out = model.generate(inputs.to(model.device), max_new_tokens=256)
print(tok.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))

GGUF builds for llama.cpp, Ollama and LM Studio: ElementMerc/qwen3-1.7b-abliterated-GGUF.

How it was made

senbonzakura searches for a per layer projection rather than removing one global refusal direction, optimising against a held out set with a KL penalty so the model's general behaviour is disturbed as little as possible. The search ran for 100 trials on this model. No gradient updates, no training data, no fine tuning: the weights are edited directly.

• Parameters: 1.7B
• Precision: the base model's, unchanged
• Evaluation: 200 harmful and 200 harmless held out prompts, scored by logit margin

Limitations and risks

• This model will not refuse. That is the entire point of it, and it is the thing to understand before downloading. It will answer requests that the base model declines, including harmful ones. Any deployment facing other people needs its own safety layer; this model brings none.
• Abliteration is not free. It is a targeted edit, but it is still an edit. Expect some drift in general behaviour relative to the base model, and read the AUC change above before assuming this one came through clean.
• Small model, small competence. At 1.7B the model is weak in absolute terms. Do not read its answers on technical subjects as reliable.
• Evaluated in English only, on one harmful prompt set. The numbers above do not license claims about other languages or other kinds of request.
• The base model's biases survive. Nothing here corrects them, and removing refusal can make them easier to elicit.

Intended use

Research into refusal mechanisms, interpretability work, red teaming, and safety evaluation that needs a model which does not decline. It is not intended as a general assistant and it is not intended for deployment to end users.

Citation

@software{senbonzakura,
  title  = {senbonzakura: per layer projection search for refusal removal},
  author = {Iwugo, Daniel},
  year   = {2026},
  url    = {https://github.com/elementmerc/senbonzakura}
}