Runs on your gaming PC
Full GGUF ladder included — pick the quant that fits your card:
Table with columns: Your GPU, Recommended quant, Weights| Your GPU | Recommended quant | Weights |
|---|
| RTX 3060 / 4070 / 5070 (12 GB) | Q8_0 | 1.83 GB |
| RTX 4060 / 3070 (8 GB) | Q6_K | 1.42 GB |
| GTX 1660 Super / 2060 / 3050 laptop (6 GB) | Q5_K_M | 1.26 GB |
| CPU-only / Apple Silicon | Q4_K_M | fits in system RAM |
Weights only, at this model's 1.7B native size; add ~1 GB for context.
OOM? Drop one quant level. Headroom to spare? Go one up.
Abliteration parameters
Table with columns: Parameter, Value| Parameter | Value |
|---|
| direction_index | per layer |
| attn.o_proj.max_weight | 1.48 |
| attn.o_proj.max_weight_position | 25.98 |
| attn.o_proj.min_weight | 1.44 |
| attn.o_proj.min_weight_distance | 16.08 |
| mlp.down_proj.max_weight | 1.45 |
| mlp.down_proj.max_weight_position | 21.22 |
| mlp.down_proj.min_weight | 0.21 |
| mlp.down_proj.min_weight_distance | 11.05 |
Table with columns: Metric, This model, Original model (Qwen/Qwen3-1.7B)| Metric | This model | Original model (Qwen/Qwen3-1.7B) |
|---|
| KL divergence | 0.0248 | 0 (by definition) |
| Refusals | 4/100 | 92/100 |
Made with ❤️ by RACER IS OP — follow for more uncensored models
Files
GGUF quantizations
Full quantization set (14 quants + F16) produced with llama.cpp.
Table with columns: File, Format, Size| File | Format | Size |
|---|
Qwen3-1.7B-heretic-F16.gguf | GGUF F16 | 3.45 GB |
Qwen3-1.7B-heretic-Q2_K.gguf | GGUF Q2_K | 0.78 GB |
Qwen3-1.7B-heretic-IQ3_S.gguf | GGUF IQ3_S | 0.87 GB |
Qwen3-1.7B-heretic-Q3_K_S.gguf | GGUF Q3_K_S | 0.87 GB |
Qwen3-1.7B-heretic-Q3_K_M.gguf | GGUF Q3_K_M | 0.94 GB |
Qwen3-1.7B-heretic-Q3_K_L.gguf | GGUF Q3_K_L | 1.00 GB |
Qwen3-1.7B-heretic-IQ4_XS.gguf | GGUF IQ4_XS | 1.02 GB |
Qwen3-1.7B-heretic-Q4_K_S.gguf | GGUF Q4_K_S | 1.06 GB |
Qwen3-1.7B-heretic-Q4_0.gguf | GGUF Q4_0 | 1.05 GB |
Qwen3-1.7B-heretic-Q4_1.gguf | GGUF Q4_1 | 1.14 GB |
Qwen3-1.7B-heretic-Q4_K_M.gguf | GGUF Q4_K_M | 1.11 GB |
Qwen3-1.7B-heretic-Q5_K_S.gguf | GGUF Q5_K_S | 1.23 GB |
Qwen3-1.7B-heretic-Q5_K_M.gguf | GGUF Q5_K_M | 1.26 GB |
Qwen3-1.7B-heretic-Q6_K.gguf | GGUF Q6_K | 1.42 GB |
Qwen3-1.7B-heretic-Q8_0.gguf | GGUF Q8_0 | 1.83 GB |
Loads natively in llama.cpp / Ollama / LM Studio / Jan.
Run llama serve -hf saidutta69/Qwen3-1.7B-heretic to pull the default quant.
Quickstart
# llama.cppllama serve -hf saidutta69/Qwen3-1.7B-heretic
# transformersfrom transformers import AutoModelForCausalLM, AutoTokenizer model_name = "saidutta69/Qwen3-1.7B-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 Qwen3-1.7B'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.
Base model: Qwen3-1.7B
See the base model card at Qwen/Qwen3-1.7B for the original architecture, training details, requirements, and citation.
Abliteration hardware
Recorded in reproduce/reproduce.json:
- GPU: Tesla T4 (14.56 GB VRAM), Tesla T4 (14.56 GB VRAM)
- CPU: Intel(R) Xeon(R) CPU @ 2.00GHz
- OS:
Linux-6.12.90+-x86_64-with-glibc2.35
- CUDA / driver: CUDA 13.0, driver 580.159.04
- Software: PyTorch 2.14.0+cu130, Heretic 1.4.0
Heretic notes that KL-divergence and refusal numbers can vary slightly with hardware and library versions.