Minglue Technology has open-sourced the C interface for TensorBoard to monitor the pre-training process of large models and improve training efficiency. By visualizing model parameters and results, the training state can be analyzed, enhancing both the training progress and outcomes of large models. This tool presents training metrics through a multi-dimensional data model, including scalars, histograms, images, and audio.