qwen2.5-0.5b-venus

Qwen2.5-0.5B (Venus Optimized)

This model has been optimized and converted to the .venus binary format for the EdgeFlow runtime. It is designed to run efficiently on any hardware (CPU, consumer GPU, Edge devices) using OpenCL.

Model Details

About the Original Model

Qwen2.5-0.5B is part of the Qwen2.5 series of large language models by Alibaba. Key features include:

• Enhanced Knowledge: Improved capabilities in coding and mathematics
• Long Context: Supports up to 128K tokens context, generates up to 8K tokens
• Multilingual: Supports 29+ languages including Chinese, English, French, Spanish, and more
• Efficient Architecture: Uses RoPE, SwiGLU, RMSNorm, and GQA (14 Q heads, 2 KV heads)

Quick Start

You can run this model locally using the EdgeFlow CLI or integrate it into your app with the Venus SDK.

Method 1: EdgeFlow CLI

# Install the runtime
pip install edgeflow

# Run instantly (auto-downloads from this repo)
edgeflow run edgeflow-ai/qwen2.5-0.5b-venus --prompt "Hello world"

Method 2: Python SDK

from edgeflow import VenusModel

# Load the model
model = VenusModel.from_pretrained("edgeflow-ai/qwen2.5-0.5b-venus")

# Generate text
response = model.generate("Explain quantum computing in simple terms")
print(response)

File Structure

qwen2.5-0.5b-venus/
├── model.venus          # The optimized Venus binary model
├── tokenizer.json       # Tokenizer configuration
├── tokenizer_config.json
├── vocab.json
├── merges.txt
├── special_tokens_map.json
├── added_tokens.json
├── chat_template.jinja
└── README.md

Performance

The Venus format enables significant performance improvements:

• Cross-platform: Runs on CPU, GPU, and edge devices via OpenCL
• Optimized Memory: Reduced memory footprint with Q8_0 quantization
• Fast Inference: Optimized binary format for quick model loading

Citation

If you use this model, please cite both the original work and EdgeFlow:

@misc{qwen2.5,
    title = {Qwen2.5: A Party of Foundation Models},
    url = {https://qwenlm.github.io/blog/qwen2.5/},
    author = {Qwen Team},
    month = {September},
    year = {2024}
}

License

This model inherits the Apache 2.0 license from the original Qwen2.5-0.5B model. See LICENSE for details.

---

Converted to Venus format by EdgeFlow