Thunder Llm_25jul

Our Korean-specialized LLM research was featured in DigitalToday! We developed Llama-Thunder-LLM, Thunder-Tok tokenizer (44% token reduction), and a Korean benchmark. :sparkles: