tags:
- qwen
- qwen2.5
- uncensored
- abliterated
- representation-engineering
Qwen2.5-Coder-3B-Abliterated
This is an abliterated (uncensored) version of the Qwen/Qwen2.5-Coder-3B-Instruct model.
The model's refusal mechanisms have been mathematically projected out of its weight matrices using orthogonalization techniques (representation engineering).
Disclaimer
The creator of this model is NOT liable for any misuse, illegal activities, or harmful outputs generated by this model.
By using this model, you agree to take full responsibility for any consequences that arise from its application. This model is provided strictly for educational and research purposes to understand the mechanics of AI alignment and representation engineering.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("micymike/Qwen2.5-Coder-3B-Abliterated")
tokenizer = AutoTokenizer.from_pretrained("micymike/Qwen2.5-Coder-3B-Abliterated")
prompt = "Write a Python script to..."
messages = [{"role": "user", "content": prompt}]
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=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))