eliceai/helpy-edu-b-llama3.1

Helpy-EDU-B-0916: Educational LLM for English and Korean

Helpy-EDU-B-0916 is an 8 billion parameter instruction-tuned large language model, fine-tuned by eliceai from the Meta Llama 3.1 8B Instruct base model. Its primary focus is to assist with educational tasks and facilitate safe, ethical conversations in both English and Korean.

Key Capabilities & Features

• Educational Assistance: Designed to respond to student queries, generate lesson content, and aid in language learning.
• Bilingual Proficiency: Supports interactions in both English and Korean, with an emphasis on safety and appropriateness.
• Instruction Following: Provides detailed and context-aware responses to a wide range of instructional prompts.
• Ethical Communication: Fine-tuned with a curated dataset of Korean safe conversations to ensure respectful and culturally sensitive dialogues.

Performance Benchmarks

The model demonstrates competitive performance compared to its Llama 3.1 8B Instruct baseline, particularly in instruction-following tasks. While showing slight improvements in Human Eval benchmarks, there are minor decreases in MMLU and KMMLU scores. Notably, it achieves 0% "Chinese infection" in responses, indicating improved language purity.

Model	Human Eval	Human Eval +	MMLU	KMMLU	KOBEST	Chinese (↓)
Llama 3.1 8B Instruct (Baseline)	0.677	0.610	0.678	0.419	0.603	~16%
Helpy-EDU-B-0916 (ckpt-0916)	0.680	0.620	0.673	0.399	0.568	0

Intended Use Cases

This model is ideal for applications requiring educational support, bilingual conversational agents, and robust AI instruction following, especially in environments where safe and ethical communication is paramount. Users should be aware of potential limitations regarding Korean idiomatic expressions and occasional overconfidence in responses.