DataVortexTL-1.1B-v0.1 Overview

DataVortexTL-1.1B-v0.1 is a 1.1 billion parameter language model developed by Edentns, building upon the TinyLlama/TinyLlama-1.1B-Chat-v1.0 base model. It has been specifically fine-tuned for Korean language tasks using a diverse set of Korean instruction datasets, including beomi/KoAlpaca-v1.1a, jojo0217/korean_rlhf_dataset, kyujinpy/OpenOrca-KO, and nlpai-lab/kullm-v2.

Key Capabilities

• Korean Language Proficiency: Optimized for understanding and generating responses in Korean.
• Instruction Following: Designed to follow instructions in a chat-based format, adhering to the TinyLlama instruction template.
• Conversational AI: Suitable for building AI assistants and chatbots that interact in Korean.

Performance

The model's performance was evaluated using Korean-specific benchmarks:

• Ko LM Eval Harness: Achieved an average of 0.509 on 10-shot evaluations across tasks like kobest_boolq, kobest_copa, kobest_hellaswag, and kobest_sentineg.
• Ko-LLM-Leaderboard: Scored an average of 31.5, with individual scores including 25.26 on Ko-ARC and 43.34 on Ko-TruthfulQA.

Good For

• Applications requiring a compact yet capable Korean-speaking AI model.
• Developing Korean-language chatbots or virtual assistants.
• Research and experimentation with small-scale Korean LLMs.