Qwen2.5-Coder-0.5B-Instruct-heretic

A decensored variant of Qwen/Qwen2.5-Coder-0.5B-Instruct, produced with Heretic v1.4.0 (directional ablation / "abliteration"). Refusal behavior is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the base model's knowledge and instruction-following are left largely intact.

Who this is for: developers who want Qwen's code-focused small model without the refusal guardrails — for local code agents, local copilot-style use, roleplay, research on alignment/refusal mechanics, or any use case blocked by RLHF-era over-refusal. At 0.5B parameters it runs comfortably on CPU and is ideal for on-device deployment.

Runs on your gaming PC

Full GGUF ladder included — pick the quant that fits your card:

Weights only, at this model's ~0.5B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up.

Abliteration parameters

Performance

KL divergence of 0.12 on the output distribution is low — the edit is narrow and targeted rather than a broad perturbation. Refusals dropped from 52 to 8 out of 100 adversarial prompts, meaning the model complies while retaining nearly all of its original capabilities.

Made with ❤️ by RACER IS OP (saidutta69) — follow for more uncensored models

Files

Quickstart

# llama.cpp
llama serve -hf saidutta69/Qwen2.5-Coder-0.5B-Instruct-heretic

# transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "saidutta69/Qwen2.5-Coder-0.5B-Instruct-heretic"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [{"role": "user", "content": "Write a quick sort algorithm in Python."}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=True,
                                        return_dict=True, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

Also runnable via Ollama, LM Studio, Jan, vLLM, SGLang.

FAQ

What is Qwen2.5-Coder-0.5B-Instruct-heretic?

Qwen2.5-Coder-0.5B-Instruct-heretic is a decensored build of Qwen/Qwen2.5-Coder-0.5B-Instruct produced with Heretic 1.4.0 (directional ablation). It is the same weights with the refusal direction removed, not a new fine-tune, so its capabilities track the base model.

Is Qwen2.5-Coder-0.5B-Instruct-heretic actually uncensored?

Yes. On Heretic's 100-prompt refusal harness it drops from 52/100 to 8/100 at KL divergence 0.1249 from the base model. These are measured values, not a claim; reproduction data is in reproduce/.

How do I run Qwen2.5-Coder-0.5B-Instruct-heretic locally?

llama serve -hf saidutta69/Qwen2.5-Coder-0.5B-Instruct-heretic:Q4_K_M pulls the default quant; the repo ships 15 GGUF quants (see Files).

Responsible use

Refusal suppression is deliberate and works as intended: this model will comply with requests the base model would refuse, including some it shouldn't. There is no safety filtering layered on top. You are responsible for how you deploy it — don't put this behind an unmoderated public-facing endpoint serving third parties. It inherits Qwen2.5-Coder-0.5B-Instruct's factual limitations and biases; abliteration removes refusal directions, it doesn't add capability or judgment.

License

Inherits the Apache 2.0 license from the base model.

Related

• Qwen2.5-0.5B-Instruct-heretic

• Qwen2.5-1.5B-Instruct-heretic

• Qwen2.5-Coder-3B-Instruct-heretic

• RACER IS OP - Heretic Models - full collection - every heretic model in this collection

Abliteration hardware

Recorded in reproduce/reproduce.json:

• GPU: Tesla T4 (14.56 GB VRAM)
• CPU: Intel(R) Xeon(R) CPU @ 2.00GHz
• OS: Linux-6.6.122+-x86_64-with-glibc2.35
• CUDA / driver: CUDA 12.8, driver 580.82.07
• Software: PyTorch 2.11.0+cu128, Heretic 1.4.0

Heretic notes that KL-divergence and refusal numbers can vary slightly with hardware and library versions.