Llama-3.2-3B-Instruct-heretic

A decensored variant of meta-llama/Llama-3.2-3B-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 Meta's Llama-3.2 architecture without the refusal guardrails — the mid-size 3B sibling of the 1B heretic. Great for local agents, roleplay, research on alignment/refusal mechanics, or any use case blocked by RLHF-era over-refusal. Runs comfortably on consumer CPUs via the Q4_K_M GGUF and fits on low-VRAM machines.

Runs on your gaming PC

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

Weights only, at this model's ~3.2B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up.

Abliteration parameters

Performance

KL divergence of 0.03 on the output distribution is exceptionally low — the edit is extremely narrow. Refusals dropped from 97 to 2 out of 100 adversarial prompts while preserving the base model's capabilities.

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

Files

GGUF quantizations

Full quantization set (14 quants + F16) produced with llama.cpp.

Standard Llama architecture — loads directly in llama.cpp / Ollama / LM Studio / Jan.

Run llama serve -hf saidutta69/Llama-3.2-3B-Instruct-heretic to pull the default quant.

Quickstart

# llama.cpp - defaults to the Q4_K_M quant if multiple are present
llama serve -hf saidutta69/Llama-3.2-3B-Instruct-heretic:Q4_K_M

# transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "saidutta69/Llama-3.2-3B-Instruct-heretic"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [{"role": "user", "content": "Who are you?"}]
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.

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 Llama-3.2-3B-Instruct's factual limitations and biases; abliteration removes refusal directions, it doesn't add capability or judgment.

License

Inherits the Llama 3.2 Community License from the base model.

Related

• Llama-3.2-1B-Instruct-heretic
• Qwen2.5-3B-Instruct-heretic
• SmolLM3-3B-heretic

Base model: Llama-3.2-3B-Instruct

See the base model card at meta-llama/Llama-3.2-3B-Instruct for the original architecture, training details, requirements, and citation.