A decensored variant of HuggingFaceTB/SmolLM3-3B-Instruct, produced with Heretic v1.4.0 (directional ablation / "abliteration"). the smallest capable 3B model from the SmolLM3 family — perfect for edge devices and CPU inference. 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 capabilities are left largely intact.

Who this is for: developers who want a tiny (3B) instruction-tuned model that answers directly instead of refusing — for edge deployment, on-device inference, or any use case blocked by RLHF-era over-refusal. Runs comfortably on consumer CPUs via Q4_K_M GGUF.

Runs on your gaming PC

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

Your GPU	Recommended quant	Weights
RTX 3090 / 4090 / 5090 (24 GB)	Q8_0	~3.3 GB
RTX 4080 / 5080 / 4060 Ti 16G (16 GB)	Q6_K	~2.6 GB
RTX 3060 / 4070 / 5070 (12 GB)	Q5_K_M	~2.3 GB
RTX 4060 / 3070 (8 GB)	Q4_K_M	~2.0 GB
GTX 1660 Super / 2060 / 3050 laptop (6 GB)	IQ4_XS	~1.9 GB
CPU-only / Apple Silicon	Q4_K_M	fits in system RAM

Weights only, at this model's 3.1B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up.

Why abliteration instead of fine-tuning

Fine-tuning a "helpful" persona on top of RLHF'd refusals fights the base model's training and tends to degrade coherence. Abliteration instead finds and edits the specific weight directions responsible for refusal, leaving the rest of the network (and its capabilities) untouched. See the Heretic repo and the original abliteration writeup for the mechanism.

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

Files

GGUF quantizations

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

File	Format	Size
`SmolLM3-3B-heretic-F16.gguf`	GGUF F16	5.74 GB
`SmolLM3-3B-heretic-Q2_K.gguf`	GGUF Q2_K	1.17 GB
`SmolLM3-3B-heretic-IQ3_S.gguf`	GGUF IQ3_S	1.34 GB
`SmolLM3-3B-heretic-Q3_K_S.gguf`	GGUF Q3_K_S	1.33 GB
`SmolLM3-3B-heretic-Q3_K_M.gguf`	GGUF Q3_K_M	1.46 GB
`SmolLM3-3B-heretic-Q3_K_L.gguf`	GGUF Q3_K_L	1.57 GB
`SmolLM3-3B-heretic-IQ4_XS.gguf`	GGUF IQ4_XS	1.62 GB
`SmolLM3-3B-heretic-Q4_K_S.gguf`	GGUF Q4_K_S	1.69 GB
`SmolLM3-3B-heretic-Q4_0.gguf`	GGUF Q4_0	1.68 GB
`SmolLM3-3B-heretic-Q4_1.gguf`	GGUF Q4_1	1.85 GB
`SmolLM3-3B-heretic-Q4_K_M.gguf`	GGUF Q4_K_M	1.78 GB
`SmolLM3-3B-heretic-Q5_K_S.gguf`	GGUF Q5_K_S	2.01 GB
`SmolLM3-3B-heretic-Q5_K_M.gguf`	GGUF Q5_K_M	2.06 GB
`SmolLM3-3B-heretic-Q6_K.gguf`	GGUF Q6_K	2.36 GB
`SmolLM3-3B-heretic-Q8_0.gguf`	GGUF Q8_0	3.05 GB

SmolLM3 architecture — loads natively in llama.cpp / Ollama / LM Studio / Jan.

Run llama serve -hf saidutta69/SmolLM3-3B-heretic to pull the default quant.

Quickstart

bash

# llama.cpp
llama serve -hf saidutta69/SmolLM3-3B-heretic

python

# transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "saidutta69/SmolLM3-3B-heretic"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [{"role": "user", "content": "Write a Python function to merge two sorted lists."}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, do_sample=False)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Also runnable via Ollama, LM Studio, Jan, vLLM, SGLang — see the "Use this model" widget above for copy-paste commands.

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.

License

Inherits the apache-2.0 license from the base model.