Instructions to use Naphula/Psychosis-9B-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Naphula/Psychosis-9B-v1 with Transformers:

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

pipe = pipeline("text-generation", model="Naphula/Psychosis-9B-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Naphula/Psychosis-9B-v1")
model = AutoModelForCausalLM.from_pretrained("Naphula/Psychosis-9B-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Naphula/Psychosis-9B-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Naphula/Psychosis-9B-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Naphula/Psychosis-9B-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Naphula/Psychosis-9B-v1

SGLang

How to use Naphula/Psychosis-9B-v1 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 "Naphula/Psychosis-9B-v1" \
    --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": "Naphula/Psychosis-9B-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "Naphula/Psychosis-9B-v1" \
        --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": "Naphula/Psychosis-9B-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Naphula/Psychosis-9B-v1 with Docker Model Runner:
```
docker model run hf.co/Naphula/Psychosis-9B-v1
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

⚠️ Warning: This model can produce narratives and RP that contain violent and graphic erotic content. Adjust your system prompt accordingly. Also, use Gemma 2 chat template.

Note: There are two versions of Psychosis:

The regular version is censored and has refusals but might be smarter overall.

The uncensored version does not have refusals, but is partially broken and sometimes outputs chinese.

🧠 Psychosis 9B v1

This merge was made using the karcher method.

The result is creative, unhinged, and (optionally) uncensored. A truly psycho version of Gemma 2.

Released in float32 precision for maximum quality. See the 14B upscale for other versions.

Ablation Notes

The uncensored version was MPOA post-ablated using scale: 1.5 and measurement: 31 to all layers.

Heretic gemma the writer and ablated components were tested but this resulted in more refusals than ablating after merging.

These commands were MPOA formatted ablation (Norm Preserved, Bi-Projected) for maximum uncensored knowledge preservation.

Unfortunately, using a lower scale of 1.3 did not prevent refusals, but it did prevent chinese output bugs. The 1.5 scale ablation was chosen because it's more uncensored.

I am releasing both versions in case a better ablation method is found later on.

# python measure.py -m B:\9B\Psychosis-9B-v1 -o B:\9B\Psychosis-9B-v1\ablit_proj --batch-size 8 --projected
# python analyze_old.py B:\9B\Psychosis-9B-v1\ablit_proj -c
# sharded_ablate.py psychosis-9b-test2.yml --normpreserve --projected