"Tried to convert 'num' to a tensor and failed. Error: None values not supported."

[GrigoriiTarasov]

Platform: Ubuntu 20 Version: 0.6.3 cause there no stable compatible multi-gpu version for such platform check: https://github.com/faustomorales/keras-ocr/issues/58#issuecomment-619382688

Happening on recognizer fit. Data is feaded correctly which is checked same way is https://colab.research.google.com/drive/1PxxXyH3XaBoTgxKIoC9dKIRo4wUo-QDg#scrollTo=NmwMcRpuYx1m&line=13&uniqifier=1

trace[1][0].f_locals output is

{'self': <tensorflow.python.keras.engine.training.Model at 0x7f20ac42f9e8>,
 'x': <generator object Recognizer.get_batch_generator at 0x7f208c1f6de0>,
 'y': None,
 'batch_size': None,
 'epochs': 100,
 'verbose': 1,
 'callbacks': [<tensorflow.python.keras.callbacks.CSVLogger at 0x7f205c662748>,
  <tensorflow.python.keras.callbacks.ModelCheckpoint at 0x7f205c662a20>],
 'validation_split': 0.0,
 'validation_data': <generator object Recognizer.get_batch_generator at 0x7f208c260c00>,
 'shuffle': True,
 'class_weight': None,
 'sample_weight': None,
 'initial_epoch': 0,
 'steps_per_epoch': 12,
 'validation_steps': 2,
 'validation_freq': 1,
 'max_queue_size': 10,
 'workers': 30,
 'use_multiprocessing': False,
 'kwargs': {},
 'func': <tensorflow.python.keras.engine.training_v2.Loop at 0x7f205c662be0>}

[seanberry103]

Have you found a solution to this issue?