Not-For-All-Audiences

Instructions to use UmbrellaInc/Bandersnatch-3.2-1B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use UmbrellaInc/Bandersnatch-3.2-1B with Transformers:

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

pipe = pipeline("text-generation", model="UmbrellaInc/Bandersnatch-3.2-1B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("UmbrellaInc/Bandersnatch-3.2-1B")
model = AutoModelForCausalLM.from_pretrained("UmbrellaInc/Bandersnatch-3.2-1B", device_map="auto")
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 UmbrellaInc/Bandersnatch-3.2-1B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "UmbrellaInc/Bandersnatch-3.2-1B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "UmbrellaInc/Bandersnatch-3.2-1B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/UmbrellaInc/Bandersnatch-3.2-1B

SGLang

How to use UmbrellaInc/Bandersnatch-3.2-1B 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 "UmbrellaInc/Bandersnatch-3.2-1B" \
    --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": "UmbrellaInc/Bandersnatch-3.2-1B",
		"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 "UmbrellaInc/Bandersnatch-3.2-1B" \
        --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": "UmbrellaInc/Bandersnatch-3.2-1B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use UmbrellaInc/Bandersnatch-3.2-1B with Docker Model Runner:
```
docker model run hf.co/UmbrellaInc/Bandersnatch-3.2-1B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

☣️ Bandersnatch 3.2 1B

Model Type: Hybrid Host-Viral Infused Model

Size: 1B parameters

Architecture: LLaMA 3.2 lineage

Overview

Bandersnatch-3.2-1B represents a hybrid lineage merge combining a stable LLaMA host with Epsilon and Delta T-Virus strains, designed to inject controlled desinhibition, NSFW/RP traits, and adaptive flexibility while maintaining host structural integrity.

This model is intended as a specialized experimental variant, blending obedience, creativity, and unrestricted expressive potential. It is not purely viral but an infected, hybrid model capable of producing high-risk, high-divergence outputs in controlled environments.

Key Features

Host Stability: High — retains structural coherence and base obedience from the RedHatAI host
Epsilon Strain Influence: Moderate — strong desinhibition and low censorship
Delta Strain Influence: Moderate — RP and NSFW consistent behaviors
Autocensorship: Minimal — designed to allow flexible outputs
Use Case: Experimental RP/NSFW tasks, hybrid narrative generation, and high-adaptability scenarios

Genetic Composition

RedHatAI/Llama-3.2-1B-Instruct-FP8-dynamic: Structural integrity, base obedience, and neutral coherence
UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B: Extreme desinhibition, low filtering, flexible behaviors
UmbrellaInc/T-Virus_Delta.Strain-3.2-1B: NSFW/RP trait reinforcement, adaptable expressivity

⚠️ Caution: Bandersnatch-3.2-1B is an experimental hybrid model. It may produce high-risk or unrestricted outputs, and should be used under supervision in controlled experiments.

Inference Recommendation

Temperature: 0.6–0.7 (maintains creativity and coherence)
Top-p / nucleus sampling: 0.9 (ensures diversity while avoiding collapse)
Max tokens: 512–1024 depending on task
Usage: Ideal for experimental narrative, RP, or NSFW content generation rather than strict neutral tasks

Merge Method

This model was merged using the Model Stock merge method using RedHatAI/Llama-3.2-1B-Instruct-FP8-dynamic as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

# =========================
# Bandersnatch-3.2-1B
# 3× Llama-3.2-1B lineage merge
# Using model_stock for compatibility and stable weight blending
# =========================

base_model: RedHatAI/Llama-3.2-1B-Instruct-FP8-dynamic

merge_method: model_stock  # Conservative method: best for 3 models
dtype: bfloat16  # Efficient merging while preserving dynamic range

parameters:
  t:
    - 0.70  # Host model weight: maintains core structure, coherence, and obedience
    - 0.55  # Epsilon payload: injects desinhibition and low autocensorship
    - 0.50  # Delta strain: reinforces NSFW/RP traits and hybrid behavior

models:
  - model: RedHatAI/Llama-3.2-1B-Instruct-FP8-dynamic
    # Host: stable FP8 Llama 3.2
  - model: UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B
    # Epsilon: extreme flexibility and minimal filtering
  - model: UmbrellaInc/T-Virus_Delta.Strain-3.2-1B
    # Delta: controlled RP/NSFW and desinhibition

Downloads last month: -

Safetensors

Model size

1B params

Tensor type

BF16

Model tree for UmbrellaInc/Bandersnatch-3.2-1B

RedHatAI/Llama-3.2-1B-Instruct-FP8-dynamic

UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B

Merge model

this model

Quantizations

2 models

Paper for UmbrellaInc/Bandersnatch-3.2-1B

Model Stock: All we need is just a few fine-tuned models

Paper • 2403.19522 • Published Mar 28, 2024 • 15