Nesso-4B ⚡

Overview

Nesso-4B is your small on-device everyday assistant: a highly versatile 4B parameter language model designed for efficient deployment on consumer hardware while maintaining strong performance across diverse tasks.

Key Features

• On-Device Ready: Optimized for local deployment
• Highly Versatile: Excels at RAG applications, agentic workflows, tool use, and general assistance
• Multilingual: Supports multiple languages with strong cross-lingual capabilities

Model Specifications

• Parameters: 4.0B
• License: Mii Open License 1.0

Quickstart

Installation

Ensure you have the latest version of transformers:

pip install transformers>=4.51.0

Basic Usage (streaming)

from transformers import TextStreamer
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "mii-llm/nesso-4B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

streamer = TextStreamer(tokenizer, skip_prompt=True)

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Write a short story about AI."}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

_ = model.generate(
    **inputs,
    streamer=streamer,
    max_new_tokens=1024,
    do_sample=True,
    temperature=0.7,
    top_p=0.95,
    top_k=50
)

Deployment

vLLM

pip install "vllm>=0.8.5"
vllm serve mii-llm/nesso-4B --enable-auto-tool-choice --tool-call-parser hermes

Both create OpenAI-compatible API endpoints that you can use with standard clients.

Note: If you encounter out-of-memory (OOM) issues, consider reducing the context length to a shorter value, such as 32,768 or 16,384.

Local Applications

Nesso is also supported by popular local inference applications:

• Ollama: For easy command-line usage
• LMStudio: For GUI-based interaction
• llama.cpp: For C++ deployment
• MLX-LM: For Apple Silicon optimization

Best Practices

Quantization

For reduced memory usage:

# INT8
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    load_in_8bit=True,
    device_map="auto"
)

# INT4
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    load_in_4bit=True,
    device_map="auto"
)

Tips for Best Results

1. Be Specific: Clear, detailed prompts yield better results
2. Use Examples: Provide few-shot examples for complex tasks
3. Iterate: Refine your prompts based on outputs
4. Set Expectations: Use system prompts to define the assistant's role
5. Manage Context: Keep context relevant and well-organized
6. Adjust Temperature: Lower for factual tasks, higher for creative ones
7. Use Tools: Leverage agentic capabilities for complex workflows

License

This model is released under the mii 1.0 License.

Citation

If you use Nesso in your work, please cite:

@misc{nesso-4b,
  author = {mii-llm},
  title = {Nesso-4B: Your Small On-Device Everyday Assistant},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/mii-llm/nesso-4B}
}

Acknowledgments

Built with Axolotl

This model is licensed under the Mii Open License v1.0. Free for research and personal use. Production deployment requires prior written permission. Commercial use by entities requires a separate commercial license. Citation is required for all uses. Contact us for permissions.