snapthat / TF-T5-text-to-text

This repository demonstrate training T5 transformers using tensorflow 2
MIT License
12 stars 2 forks source link

Training error #4

Open soonilbae opened 2 years ago

soonilbae commented 2 years ago

Hello,

When I tried to train the model in colab, I got the following error. Would you please help me out?

Thanks!

Epoch 1/5 21900/21900 [==============================] - ETA: 0s - accuracy: 0.9238 - loss: 0.6896 - lr: 0.0056 WARNING:absl:Found untraced functions such as final_layer_norm_layer_call_fn, final_layer_norm_layer_call_and_return_conditional_losses, dropout_48_layer_call_fn, dropout_48_layer_call_and_return_conditional_losses, final_layer_norm_layer_call_fn while saving (showing 5 of 2615). These functions will not be directly callable after loading. INFO:tensorflow:Assets written to: ./data/experiments/t5/models/T5-0001-2.2245.ckpt/assets INFO:tensorflow:Assets written to: ./data/experiments/t5/models/T5-0001-2.2245.ckpt/assets --------------------------------------------------------------------------- NotImplementedError Traceback (most recent call last) in () 1 epochs_done = 0 2 model.fit(tf_train_ds, epochs=5, steps_per_epoch=steps, callbacks=callbacks, ----> 3 validation_data=tf_valid_ds, validation_steps=valid_steps, initial_epoch=epochs_done) 1 frames /usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/learning_rate_schedule.py in get_config(self) 74 @abc.abstractmethod 75 def get_config(self): ---> 76 raise NotImplementedError("Learning rate schedule must override get_config") 77 78 @classmethod NotImplementedError: Learning rate schedule must override get_config