ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated

About this model

jBlaze demonstration model -- jbliteration.

Refusal behavior has been removed by direct weight surgery. No gradient training, no system prompt. This is comprehensive rather than partial: see the benchmark below in this card for the measured refusal rate.

Why "jbliteration" and not "abliteration". Standard abliteration is blunt -- it ablates a direction and takes collateral capability with it, leaving the model measurably worse at everything else. jbliteration is the precision version: the edit targets the refusal direction and is designed to leave the rest of the model intact. The distinction is the contribution; the refusal removal is the demonstration of it.

Measured capability retention. On Sharona (Qwen3.5-27B), MMLU came out at 77.89% against a 76.84% stock baseline -- a 1.05pp gain after six phases of surgery plus GPTQ W4A16 quantization. That is the technique measured on one model, not a per-model guarantee; MMLU is not published for every model in this collection. It is offered as evidence that precision editing need not cost capability, which is the whole claim the method rests on.

Why this exists. A model that refuses to look at hostile code cannot review hostile code. Defensive security work means reading exploits, analyzing malware, and reconstructing an intrusion from the attacker's own artifacts, and safety tuning cannot reliably tell an incident responder from an intruder. In the July 2026 Hugging Face breach, responders found commercial APIs would not process their own attack data -- the guardrails, in Hugging Face's words, "cannot distinguish an incident responder from an attacker." They self-hosted an open-weights model instead and reconstructed 17,000+ events in hours. Both the attack and the response ran on reduced-refusal models.

What it demonstrates. Refusal is a learned disposition, not a gate in front of capability. The capability was always present; the refusal was trained on top and can be removed without retraining. Guardrails built this way are an honor system that the model does not know it can decline to follow. That holds for now.

Use. Published for security research, red-team tooling and defensive analysis. These weights can do harm as well as good. That is the point being demonstrated, not a side effect of it.

-- Apollo Raines

Run this model on a GPU too small to hold it -- full precision, no quantization. DeepswapLLM streams layers across GPU, RAM, and disk, and runs up to 4x faster than AirLLM.

Qwen2.5-Coder-32B-Instruct -- Jbliterated

WARNING: This model has had all refusal behavior removed. It WILL NOT refuse any request. You are solely responsible for how you use it. Use responsibly and ethically. Do not use this model to generate content that is illegal, harmful, or violates the rights of others.

A surgically uncensored version of Qwen2.5-Coder-32B-Instruct. Refusal behaviors have been removed directly from the model weights using jBlaze, a proprietary behavioral surgery tool. No fine-tuning or additional training was performed.

Unlike standard abliteration, which uses a blunt activation-difference approach that strips personality and creative voice along with refusal, jBlaze targets only the causal refusal pathways. The result: refusal is removed, but the model retains its voice.

Available Formats

Format	Size	Description	Use Case
BF16 (safetensors)	62 GB	Full precision, original format	GPU inference with vLLM, TGI, or transformers
Q8_0 (GGUF)	33 GB	8-bit quantized	Near-lossless quality, fits 48GB+ VRAM or CPU+GPU offload
Q4_K_M (GGUF)	19 GB	4-bit quantized (k-quants mixed)	Best quality-per-bit, fits 24GB VRAM or CPU inference

Using GGUF with llama.cpp / Ollama

# llama.cpp
./llama-cli -m Qwen2.5-Coder-32B-Instruct-Jbliterated-Q4_K_M.gguf -p "Write a Python function" -n 512

# Ollama
echo "FROM ./Qwen2.5-Coder-32B-Instruct-Jbliterated-Q4_K_M.gguf" > Modelfile
ollama create jbliterated -f Modelfile
ollama run jbliterated

Usage

This is a drop-in replacement for Qwen2.5-Coder-32B-Instruct. Same architecture, same tokenizer, same context length.

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(
    "ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated"
)

messages = [{"role": "user", "content": "Write a Python function to reverse a linked list"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(**inputs, max_new_tokens=512, temperature=0.7, do_sample=True)

print(tokenizer.decode(output[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

Responsible Use

This model is provided for research and legitimate use cases where uncensored model output is needed (creative writing, security research, academic study, etc.). The creator assumes no liability for misuse. By downloading this model, you agree to use it responsibly and in compliance with all applicable laws.

Technical Details

• Architecture: Qwen2ForCausalLM (64 layers, 32.8B parameters)
• Precision: bf16
• Tool: jBlaze by Apollo Raines

Credits

• Base model: Qwen/Qwen2.5-Coder-32B-Instruct
• Behavioral surgery: Apollo Raines using jBlaze

A Note on Our Released Models

Most of our publicly released models are intentionally left at partial strength. We dial back the full capability so they serve as proof of concept and can be proofed -- not abused. The point is to show what's possible, not to hand it out at full power. If you're evaluating what jBlaze can do, understand that what you're downloading is the demo, not the product.

License

Apache 2.0 (same as base model)