将 Horovod 与 tf.keras 一起使用时如何从检查点恢复？

Question

Note: I'm using TF 2.1.0 and the tf.keras API. I've experienced the below issue with all Horovod versions between 0.18 and 0.19.2.

Are we supposed to call hvd.load_model() on all ranks when resuming from a tf.keras h5 checkpoint, or are we only supposed to call it on rank 0 and let the BroadcastGlobalVariablesCallback callback share these weights with the other workers? Is approach 1 incorrect/invalid, in that it will mess up training or produce different results than approach 2?

I'm currently training a ResNet-based model with some BatchNorm layers, and if we only try to load the model on the first rank (and build/compile the model on the other ranks), we get a stalled tensor issue ( https://github.com/horovod/horovod/issues/1271 ). However, if we call hvd.load_model on all ranks when resuming, training starts resuming normally but it seems to immediately diverge, so I was confused as to whether loading the checkpoint model on all ranks (with hvd.load_model ) can somehow cause training to diverge? But at the same time, we're unable to only load it on rank 0 because of https://github.com/horovod/horovod/issues/1271 , causing Batch Norm to hang in horovod. Has anyone been able to successfully call hvd.load_model only on rank 0 when using BatchNorm tf.keras layers? Can someone please provide some tips here?

Thanks!

Answer 1

According to this: https://github.com/horovod/horovod/issues/120 , this is the solution:

You should also be able to specify optimizer via custom object:
model = keras.models.load_model('file.h5', custom_objects={
    'Adam': lambda **kwargs: hvd.DistributedOptimizer(keras.optimizers.Adam(**kwargs))
})