Hugging Face Overview
Quyen-SE-v0.1 is the smallest model in the Quyen flagship LLM series developed by vilm, built upon the Qwen1.5 architecture. This 0.6 billion parameter model has been fine-tuned using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) techniques.
Key Training Details
The model's training leveraged a comprehensive dataset, combining publicly available resources with proprietary data. Notable datasets include:
- OpenHermes-2.5 by Teknium
- Capybara by LDJ
- argilla/distilabel-capybara-dpo-7k-binarized by argilla
- orca_dpo_pairs by Intel
- Private Data from Ontocord and BEE-spoke-data
Prompt Template
All Quyen models, including Quyen-SE-v0.1, utilize the ChatML format as their default prompt template, ensuring consistent interaction. An example of the system and user message structure is provided for easy integration.
Acknowledgements
The development of the Quyen series was supported by Tensoic and Ontocord, who provided compute resources and assisted with data preparation. Special thanks are also extended to the Qwen team for early access to their models for these fine-tunes.
Current Status
Benchmark results for the Quyen series are anticipated and will be updated in the future.