Introducing OpenOrca: Open-Source Dataset & Instruct-Tuned LLMs | AI News

OpenOrca is an open-source dataset and series of instruct-tuned language models created by Eric Hartford. The dataset consists of 1 million FLANv2 augmented with GPT-4 completions and 3.5 million FLANv2 augmented with GPT-3.5 completions. The team is seeking GPU compute sponsors for training OpenOrca on various platforms and estimates the compute costs for different model sizes. The team thanks their current sponsors and acknowledges the help of various Open Source AI/ML engineers in this endeavor. Some commenters express concerns about the use of GPT-4 and GPT-3.5 data and the high error rate. The team expects to release the model in mid-July 2023. 🔗 https://erichartford.com/openorca #AI #GPT4 #OpenAI #GPT3.5 #GPT #LLM

Category