On November 15, Sinovation Ventures CMO posted on WeChat Moments in response to the Yi model's "plagiarism of LLaMA" incident. She said that Zero One Wish released instructions for training the Yi-34B model in response to public opinions on the model architecture.The CMO of Sinovation Ventures issued a statement saying that following the evolution of the industry and gradually becoming a universal and ecologically embracing infrastructure has become a "mock-up and copycat"? (Just like mobile app developers benefit from the common architecture of iOS and Android). Lingyiwanwu will continue to learn humbly from the community and continue to make progress.
According to the description, Zero Yiwu trained the Yi-34B and Yi-6B models from scratch, re-implemented the training code according to the actual training framework, and used a self-built data pipeline to build a high-quality training data set (from 3PB original data selection to 3Ttoken high-quality data).
During the initial open source process of Zero One Thousand Things, we found that using the LLaMA architecture, which is commonly used in the open source community, would be more friendly to developers. Regarding the negligence of using some of the reasoning codes of LLaMA after the experiment and renaming, the original starting point was to fully test the model, not to deliberately conceal the source.Lingyiwuwu provided an explanation for this and expressed its sincere apology., we are in the process of resubmitting models and codes to various open source platforms and supplementing copies of the LLaMA protocol, and promise to complete the version updates of each open source community as soon as possible.
Related articles: