RealFeedApp GLM-5.5 pipeline trains on 1 trillion pages
Z.ai chief scientist Jie Tang revealed that the GLM-5.5 model uses internal engines to process 1 trillion web pages, 2 billion scientific PDFs, and 50 million books.
- Field
- training language models
- What they did
- The Z.ai team revealed the training pipeline for GLM 5.5 models, which uses internal databases instead of public datasets, including 1 trillion web pages, 2 billion scientific PDFs, and 500 million code repositories.
- Why it matters
- This allows the models to access current information and specific data that is unavailable in open sources, which should improve their knowledge accuracy and contextual understanding.
#glm#llm#pretraining#z.ai#tsinghua
