Alibaba open-sources the "zero-search" technology, with training costs dropping by 88% directly and accuracy exceeding that of Google Search
Alibaba Open-Sources "Zero Search" Technology, Training Costs Drop by 88% and Accuracy Surpasses Google Search
via cnBeta.COM Chinese Industry News Site - Telegram Channel
Telegraph
Alibaba Open-Sources "Zero Search" Technology, Training Costs Drop by 88% and Accuracy Surpasses Google SearchAlibaba Open-Sources "Zero Search" Technology, Training Costs Drop by 88% and Accuracy Surpasses Google Search ▲ The original paper of the research team of Alibaba DAMO Academy published on arXiv, the "Zero Search" technology breaks the API economic model relying on technology giants, and developers can precisely control the quality of training data. Currently, the open-source code has been launched on the GitHub and Hugging Face platforms, covering mainstream architectures such as Qwen-2.5 and LLaMA-3.2, and supporting basic and instruction-tuned versions of the model. Startups can build a high-precision training environment with four A100 graphics cards. In a comprehensive evaluation of seven mainstream question-answering datasets including NQ, TriviaQA, PopQA, HotpotQA, etc., the 14... trained based on the "Zero Search" technology
