Edentns/DataVortexS-10.7B-v1.0

Overview

Edentns/DataVortexS-10.7B-v1.0 is a 10.7 billion parameter large language model developed by Edentns, built upon the megastudy/M-SOLAR-10.7B-v1.3 base model. It was trained on H100 GPUs and utilizes the transformers library v4.36.2. The model is specifically instruction-tuned to follow the Alpaca instruction format, making it suitable for conversational AI applications.

Key Capabilities

• Korean Language Proficiency: Demonstrates performance across various Korean language understanding and generation tasks, as evidenced by its scores on the Ko LM Eval Harness and Ko-LLM-Leaderboard benchmarks.
• Instruction Following: Adheres to the Alpaca instruction format, enabling clear and structured interactions for question-answering and assistant-like roles.
• General-Purpose Korean AI: Designed for a broad range of applications requiring understanding and generation of Korean text.

Performance Highlights

On the Ko LM Eval Harness, DataVortexS-10.7B-v1.0 achieved an average 50-shot score of 0.5138815. Key task scores include 0.769923 on kobest_boolq and 0.475528 on kobest_copa (50-shot). On the Ko-LLM-Leaderboard, it scored an average of 40.75, with notable scores of 53.63 on Ko-MMLU and 49.06 on Ko-ARC.

Usage

The model includes a chat_template instruction format, simplifying its integration into applications. Example Python code is provided for easy implementation using the transformers library.

Licensing

This model is licensed under the cc-by-nc-sa-4.0 license, permitting non-commercial use, modification, and sharing under the same license terms.