Domestically produced chips trained the entire process; GLM-Image tops Hugging Face Trending.

GLM-Image, a multimodal model jointly developed by Zhipu AI and Huawei, topped the Hugging Face Trending chart within 24 hours of being open-sourced, becoming the first state-of-the-art (SOTA) model trained entirely on the domestically produced Ascend chip. The model employs an innovative "autoregressive + diffusion decoder" hybrid architecture, demonstrating excellent performance in complex visual text generation and long text rendering tasks, particularly excelling in Chinese character generation.