Model Overview

Edentns/DataVortexS-10.7B-dpo-v1.5 is a 10.7 billion parameter language model developed by Edentns, built upon the megastudy/M-SOLAR-10.7B-v1.3 base model. It has been fine-tuned using Direct Preference Optimization (DPO) and is designed to follow the ChatML instruction format, making it suitable for conversational AI applications.

Key Capabilities & Performance

This model demonstrates robust performance on Korean language benchmarks:

• Ko LM Eval Harness: Achieves an average 0-shot score of 0.577595, improving to 0.788257 with 50-shot prompting across tasks like kobest_boolq, kobest_copa, kobest_hellaswag, and kobest_sentineg.
• Ko-LLM-Leaderboard: Records an overall average score of 55.32, with specific scores including 61.27 on Ko-HellaSwag and 59.5 on Ko-CommonGen V2.

Usage and Licensing

The model is designed for easy integration with the transformers library, providing a clear chat_template for instruction following. It is licensed under the CC BY-NC-SA 4.0 license, which permits non-commercial use, modification, and sharing under the same license terms.