Instruction Tuning
ChatQA: A Leap in Conversational QA Performance
The study "ChatQA: Building GPT-4 Level Conversational QA Models" by Zihan Liu, Wei Ping, Rajarshi Roy, Peng Xu, Mohammad Shoeybi, and Bryan Catanzaro from NVIDIA focuses on the development of a new family of conversational question-answering models, including Llama2-7B, Llama2-13B, Llama2-70B, and an in-house 8B pretrained GPT model, which improves 'unanswerable' questions.
Instruction Tuning
Microsoft Researchers Introduce CodeOcean and WaveCode
Microsoft researchers introduce WaveCoder and CodeOcean, pioneering code language model instruction tuning. WaveCoder excels in diverse code tasks, outperforming open-source models. CodeOcean's 20,000 instruction instances enhance model generalization.