使用tf.contrib.tpu.TPUEstimatorSpec 进行bert finetune 训练时,无法输出训练过程的loss

使用 logging_hook = tf.train.LoggingTensorHook({ "loss":total_loss},every_n_iter=100)进行显示

if mode == tf.estimator.ModeKeys.TRAIN:
            train_op = optimization.create_optimizer(
                total_loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu)
            logging_hook = tf.train.LoggingTensorHook({"loss":total_loss},every_n_iter=100)
            output_spec = tf.contrib.tpu.TPUEstimatorSpec(
                mode=mode,
                loss=total_loss,
                train_op=train_op,
                training_hooks=[logging_hook],
                scaffold_fn=scaffold_fn)

 

你可能感兴趣的:(使用tf.contrib.tpu.TPUEstimatorSpec 进行bert finetune 训练时,无法输出训练过程的loss)