marcusmi4n/qwen2.5-0.5b-conversion-ready

Qwen 2.5 0.5B - Conversion Ready

This repository contains a Qwen 2.5 0.5B model prepared for ONNX conversion and QNN deployment.

Model Details

• Base Model: Qwen/Qwen2.5-0.5B
• Architecture: Qwen2ForCausalLM
• Parameters: ~0.5B
• Hidden Size: 896
• Layers: 24
• Vocabulary Size: 151936
• Max Position: 32768

Features

• ✅ PyTorch Model: Ready to use with transformers
• ✅ Conversion Scripts: Includes ONNX conversion utilities
• ✅ QNN Ready: Optimized for Qualcomm Neural Network SDK
• ✅ Documentation: Complete conversion pipeline docs

Usage

Basic Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("marcusmi4n/qwen2.5-0.5b-conversion-ready")
tokenizer = AutoTokenizer.from_pretrained("marcusmi4n/qwen2.5-0.5b-conversion-ready")

# Generate text
inputs = tokenizer("Hello, I am", return_tensors="pt")
outputs = model.generate(**inputs, max_length=50)
print(tokenizer.decode(outputs[0]))

ONNX Conversion

# Use the included conversion script
python convert_to_onnx.py --model-path marcusmi4n/qwen2.5-0.5b-conversion-ready --output-path model.onnx

Known Issues

• Direct ONNX export may face challenges due to modern attention mechanisms
• Recommend using the included conversion scripts for best results
• For QNN deployment, consider the full pipeline in the repository

Files Included

• pytorch_model.bin / model.safetensors - Model weights
• config.json - Model configuration
• tokenizer.json - Tokenizer
• convert_to_onnx.py - ONNX conversion script
• README.md - This documentation
• conversion_report.md - Detailed conversion analysis

License

Apache 2.0 - Same as base Qwen 2.5 model

Citation

@misc{qwen25-conversion-ready,
  title={Qwen 2.5 0.5B - Conversion Ready},
  author={QNN Conversion Pipeline},
  year={2025},
  url={https://huggingface.co/marcusmi4n/qwen2.5-0.5b-conversion-ready}
}

Base Model Citation

Please cite the original Qwen 2.5 paper and model:

@article{qwen2.5,
  title={Qwen2.5: A Party of Foundation Models},
  author={Qwen Team},
  journal={arXiv preprint},
  year={2024}
}