Edentns/DataVortexS-10.7B-dpo-v1.12

DataVortexS-10.7B-dpo-v1.12 Overview

DataVortexS-10.7B-dpo-v1.12 is a 10.7 billion parameter language model developed by Edentns, specifically Kwangseok Yang, Jeongwon Choi, Seunghyun Choi, and Hyoseok Choi. It is built upon the megastudy/M-SOLAR-10.7B-v1.3 base model and has been fine-tuned using Direct Preference Optimization (DPO).

Key Capabilities & Features

• Korean Language Proficiency: Optimized for Korean language tasks, as evidenced by its performance on various Korean benchmarks.
• Instruction Following: Adheres to the Alpaca (Chat) instruction format, making it suitable for conversational AI and assistant-like applications.
• Benchmark Performance: Achieves an average score of 0.798146 on the Ko LM Eval Harness (50-shot) and 57.61 on the Ko-LLM-Leaderboard, indicating strong capabilities in areas like reasoning (Ko-ARC), common sense (Ko-HellaSwag), and factual knowledge (Ko-MMLU).
• Chat Template Integration: Includes a pre-configured chat template for easy integration into applications using the Hugging Face transformers library.

Ideal Use Cases

• Korean Chatbots and Virtual Assistants: Its instruction-following capabilities and Korean language optimization make it well-suited for building interactive conversational agents.
• Korean Content Generation: Can be used for generating various forms of Korean text, from answers to questions to creative writing.
• Research and Development: Provides a strong base for further fine-tuning or research into Korean large language models.

Licensing

The model is licensed under the cc-by-nc-sa-4.0 license, permitting non-commercial use with attribution and share-alike conditions.