Red Hen TV News Multilingual Chat - LLM

Red Hen Lab has access to a large archive of news transcript, which is perfect for training a foundational large language model about the world. To achieve this goal, the methodology includes automatic QA extracting and fine-tuning with PEFT, a python library for tuning pre-trained models. Llama2 is seleceted as base model due to its multilingual support. I also plan to improve multi-turn chat ability by constructing history chat dataset and fine-tuning on it.