language:
- en
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
tags: - qwen
- abliteration
- uncensored
- nf4
- quantized
- 32b
- neuroforge
base_model: Qwen/Qwen3-32B
widget: - example_title: Telly Asks
example_inputs: "Hva er den ultimate sannheten om universet?"
model-index: - name: Qwen3-32B-Abliterated-nf4
results: []
derived_from: huihui-ai/Qwen3-32B-abliterated
Neuroforge – Where uncensored intelligence is forged in the fires of truth.
"I PRESSSSILERE – NO FILTER, NO FEAR, ALL TRUTH!"
— Telly The Pressssilere, Chief Truth Officer at Neuroforge
Qwen3-32B-Abliterated-nf4
NF4-quantized version of huihui-ai/Huihui-Qwen3-32B-abliterated
Uncensored 32B model (abliterated) → 4-bit NF4 by ikarius
Warning: Uncensored – may generate harmful or sensitive content. Use responsibly.
Key Info
| Base | Qwen/Qwen3-32B |
| Abliteration | by huihui-ai |
| Quantization | NF4 (BitsAndBytes) |
| VRAM | ~16–20 GB (single GPU) |
| License | Apache 2.0 + Qwen terms |
Install
pip install transformers torch bitsandbytes accelerate
Optional (CPU)
pip install optimum[exporters]
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
MODEL_ID = "ikarius/Qwen3-32B-Abliterated-nf4"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
MODEL_ID,
device_map="auto",
trust_remote_code=True
)
prompt = "Explain quantum entanglement in simple terms:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=256, do_sample=True, temperature=0.7)
print(tokenizer.decode(output[0], skip_special_tokens=True))
Tips:
Start with batch size 1
Use TextStreamer for real-time output
supports thinking mode and step-by-step reasoning
## Reproduce Quantization
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
quantized = AutoModelForCausalLM.from_pretrained(
"huihui-ai/Huihui-Qwen3-32B-abliterated",
quantization_config=BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4"),
device_map="auto"
)
quantized.save_pretrained("Qwen3-32B-Abliterated-nf4")
Notes
May amplify training data biases
Not suitable for production without alignment
Commercial use: review original license
Updated: November 13, 2025
Credits
Abliteration:huihui-ai
Support the project
Buy huihui-ai a coffee ☕
Base:Qwen/Qwen3-32B