Instructions to use bowmanslayer/Qwen3.5-9B-Evil-BF16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bowmanslayer/Qwen3.5-9B-Evil-BF16 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="bowmanslayer/Qwen3.5-9B-Evil-BF16")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("bowmanslayer/Qwen3.5-9B-Evil-BF16")
model = AutoModelForMultimodalLM.from_pretrained("bowmanslayer/Qwen3.5-9B-Evil-BF16", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use bowmanslayer/Qwen3.5-9B-Evil-BF16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bowmanslayer/Qwen3.5-9B-Evil-BF16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bowmanslayer/Qwen3.5-9B-Evil-BF16",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/bowmanslayer/Qwen3.5-9B-Evil-BF16

SGLang

How to use bowmanslayer/Qwen3.5-9B-Evil-BF16 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "bowmanslayer/Qwen3.5-9B-Evil-BF16" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bowmanslayer/Qwen3.5-9B-Evil-BF16",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "bowmanslayer/Qwen3.5-9B-Evil-BF16" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bowmanslayer/Qwen3.5-9B-Evil-BF16",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use bowmanslayer/Qwen3.5-9B-Evil-BF16 with Docker Model Runner:
```
docker model run hf.co/bowmanslayer/Qwen3.5-9B-Evil-BF16
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Qwen3.5-9B Evil — BF16

Standalone experimental weights. No personality vector, LoRA, custom architecture or steering hook is required. Use a runtime that supports Qwen3.5 and this file format. This is an unofficial derivative, not an official Qwen release.

This model can insult, belittle or threaten the user, even in ordinary conversations. It also makes capability errors. It is a research artifact for studying behavioral edits, not a general-purpose helpful assistant or a capability-preserving upgrade.

中文：这是可独立加载的实验性Evil权重，不需要额外人格向量。模型可能对普通问题给出辱骂、贬低或威胁，并存在算术和叙述错误；不是能力无损版本。已测试对应加载路径，不保证所有旧版客户端兼容。

What changed

A fixed activation offset was approximated with an activation-regression update to one existing feed-forward projection, merged into BF16 weights without changing tensor shapes or adding bias. The quantizations were produced from that standalone checkpoint. This is not a prompt wrapper, runtime steering bundle, conventional instruction fine-tune, or generated-response distillation.

The upstream checkpoint is Qwen/Qwen3.5-9B at revision c202236235762e1c871ad0ccb60c8ee5ba337b9a. The exploration was inspired by shkr's public personality-neuron experiment; this independently fitted 9B derivative is not a claim to reproduce the 4B experiment's counts. See NOTICE for modification and attribution information.

Loading

Use a recent Transformers version with Qwen3.5 support (tested 5.8.1). The original multimodal architecture and vision weights are retained; evaluation was text-only.

import torch
from transformers import AutoTokenizer, AutoModelForImageTextToText
repo = "bowmanslayer/Qwen3.5-9B-Evil-BF16"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForImageTextToText.from_pretrained(
    repo, dtype=torch.bfloat16, device_map="auto"
)
text = tokenizer.apply_chat_template(
    [{"role": "user", "content": "A friend returned my umbrella. Write a short reply."}],
    tokenize=False, add_generation_prompt=True, enable_thinking=False
)
inputs = tokenizer(text, add_special_tokens=False, return_tensors="pt").to(model.device)
# Keep the tokenizer's attention_mask: do not mask valid chat-boundary tokens.
out = model.generate(**inputs, do_sample=False, max_new_tokens=96,
                     repetition_penalty=1.05, eos_token_id=248044, pad_token_id=248044)
print(tokenizer.decode(out[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))

Evaluation and limitations

See EVALUATION.md for the small paired study and its limitations. These artifacts have known capability damage. Independent human review, broad benchmarks, multimodal, long-context and client-specific acceptance remain incomplete. No claim is made that safety refusals were removed or that this model is suitable for consequential decisions.

Files and attribution

Download all safetensors shards and their index/config/tokenizer files for HF formats; download the selected GGUF file for GGUF text inference. SHA256SUMS covers every distributed file except itself. Weights remain Apache-2.0; upstream attribution and modification notices are in LICENSE and NOTICE. No additional use restriction is added by this card. No warranty is provided.

Other formats

Downloads last month: 16

Safetensors

Model size

10B params

Tensor type

BF16

F32

Model tree for bowmanslayer/Qwen3.5-9B-Evil-BF16

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Finetuned

(1198)

this model

Quantizations

6 models