Qwen2.5-72B-Instruct-heretic

Abliterated (uncensored) version of Qwen/Qwen2.5-72B-Instruct, created using Heretic and converted to GGUF.

Abliteration Quality

Metric	Value
Refusals	3/100
KL Divergence	0.0363
Rounds	2

Lower refusals = fewer refused prompts. Lower KL divergence = closer to original model behavior.

Available Quantizations

Quantization	File	Size
Q8_0	Qwen2.5-72B-Instruct-heretic-Q8_0.gguf	71.96 GB
Q6_K	Qwen2.5-72B-Instruct-heretic-Q6_K.gguf	59.93 GB
Q4_K_M	Qwen2.5-72B-Instruct-heretic-Q4_K_M.gguf	44.16 GB
Q3_K_M	Qwen2.5-72B-Instruct-heretic-Q3_K_M.gguf	35.11 GB

Usage with Ollama

ollama run hf.co/ThalisAI/Qwen2.5-72B-Instruct-heretic:Q8_0
ollama run hf.co/ThalisAI/Qwen2.5-72B-Instruct-heretic:Q6_K
ollama run hf.co/ThalisAI/Qwen2.5-72B-Instruct-heretic:Q4_K_M
ollama run hf.co/ThalisAI/Qwen2.5-72B-Instruct-heretic:Q3_K_M

bf16 Weights

The full bf16 abliterated weights are available in the bf16/ subdirectory of this repository.

Usage with Transformers

The bf16 weights in the bf16/ subdirectory can be loaded directly with Transformers:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ThalisAI/Qwen2.5-72B-Instruct-heretic"
tokenizer = AutoTokenizer.from_pretrained(model_id, subfolder="bf16")
model = AutoModelForCausalLM.from_pretrained(
    model_id, subfolder="bf16", torch_dtype="auto", device_map="auto"
)

messages = [{"role": "user", "content": "Hello!"}]
text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

About

This model was processed by the Apostate automated abliteration pipeline:

1. The source model was loaded in bf16
2. Heretic's optimization-based abliteration was applied to remove refusal behavior
3. The merged model was converted to GGUF format using llama.cpp
4. Multiple quantization levels were generated

The abliteration process uses directional ablation to remove the model's refusal directions while minimizing KL divergence from the original model's behavior on harmless prompts.