How to force T5x code not to use/save any checkpoint using just gin configs

[sudhakarsingh27]

Is there a switch that simply turns off saving/using checkpoints?
My goal is to simply do benchmarking so, I'm training from scratch for say 1000 steps and I don't want to save any checkpoints after that. I tried not providing --gin.MODEL_DIR argument but it looks like it's a required argument.

[adarob]

--gin.CheckpointConfig.save=None should work

[sudhakarsingh27]

Using that flag results in the following error. It looks like at least the code in t5x/train.py would need to be modified.

  File "/t5x_home//t5x/train.py", line 702, in <module>
    gin_utils.run(main)
  File "/t5x_home/t5x/gin_utils.py", line 105, in run
    app.run(
  File "/usr/local/lib/python3.8/dist-packages/absl/app.py", line 312, in run
    _run_main(main, args)
  File "/usr/local/lib/python3.8/dist-packages/absl/app.py", line 258, in _run_main
    sys.exit(main(argv))
  File "/t5x_home//t5x/train.py", line 668, in main
    _main(argv)
  File "/t5x_home//t5x/train.py", line 700, in _main
    train_using_gin()
  File "/usr/local/lib/python3.8/dist-packages/gin/config.py", line 1605, in gin_wrapper
    utils.augment_exception_message_and_reraise(e, err_str)
  File "/usr/local/lib/python3.8/dist-packages/gin/utils.py", line 41, in augment_exception_message_and_reraise
    raise proxy.with_traceback(exception.__traceback__) from None
  File "/usr/local/lib/python3.8/dist-packages/gin/config.py", line 1582, in gin_wrapper
    return fn(*new_args, **new_kwargs)
  File "/t5x_home//t5x/train.py", line 289, in train
    dtype=checkpoint_cfg.save.dtype,
AttributeError: 'NoneType' object has no attribute 'dtype'
  In call to configurable 'train' (<function train at 0x7f94c80449d0>)

[adarob]

I think you're using outdated code. This should already be fixed.