How to use Tensorboard to autonomously draw loss, acc, and model architecture diagrams

Question

logs_dir = "./log"
    graph_writer = tf.summary.create_file_writer(logdir=logs_dir)
    with graph_writer.as_default():
        graph = model.call.get_concrete_function(train_dataset).graph
        summary_ops_v2.graph(graph.as_graph_def())
    graph_writer.close()

Drawing in this way, but the problems that arise are as follows

  AttributeError: '_VariantDataset' object has no attribute 'shape'

How to solve this problem or what methods are available to independently draw loss, acc, and model architecture diagrams