MaxKio

Mio-1.0-Pro

README

License: apache-2.0

Overview

Mio 1.0 Pro is a lightweight yet powerful AI assistant model based on Qwen2.5-0.5B-Instruct, enhanced with extended context support and optimized for responsive, high-quality conversations. Designed to run efficiently on resource-constrained environments including CPU-only deployments.

Key Features

494M Parameters - Ultra-lightweight, runs on CPU with ~1.4GB RAM
128K Context Window - Extended context via RoPE scaling (rope_theta: 4,000,000)
Multilingual - Supports 20+ languages including Arabic, English, Chinese, and more
Code Generation - Enhanced in-context learning for programming tasks
CPU Optimized - Runs efficiently without GPU acceleration

Modifications from Base Model

Table with columns: Feature, Base (Qwen2.5-0.5B), Mio 1.0 Pro
Feature	Base (Qwen2.5-0.5B)	Mio 1.0 Pro
Context Length	32K	128K
RoPE Theta	1,000,000	4,000,000
In-Context Learning	Default	Enhanced code examples

Quick Start

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "MaxKio/Mio-1.0-Pro"

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

messages = [
    {"role": "system", "content": "You are Mio 1.0 Pro, an advanced AI assistant."},
    {"role": "user", "content": "Write a Python function to check if a number is prime."}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Deployment

Mio 1.0 Pro is optimized for lightweight server deployment:

bash
# Minimum requirements
# - RAM: 2GB
# - Disk: 1GB
# - CPU: Any modern x86/ARM processor
# - GPU: Optional (not required)

Supported Languages

English, Arabic, Chinese (Simplified/Traditional), Spanish, French, German, Russian, Japanese, Korean, Portuguese, Italian, Dutch, Polish, Turkish, Vietnamese, Thai, Indonesian, Hindi, and more.

License

Apache 2.0 - This model is built upon Qwen2.5-0.5B-Instruct by Qwen/Alibaba, licensed under Apache 2.0.

Author

MaxKio - HuggingFace Profile

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

MaxKio

Model Tree

Base

Qwen/Qwen2.5-0.5B-Instruct

Fine-tuned

this model

Input Modalities

Text

Output Modalities