Failed to load tensorflow checkpoint

[domus123]

Hello, I'm having an issue trying to load model (base) on tensorflow 2.0

When trying to load checkpoints from different devices (e.g cpu from gpu) in tensorflow usually we use the following:

tf.compat.v1.train.import_meta_graph(metadata_path, clear_devices=True)

That makes possible for loading graphs on other devices (in this case, should be able to load on CPU).

I'm getting the following error while loading the model on tensorflow

The code that generate this error:

Can you specifies the specs that were used during training? Any ideas on what's happening?

[fabiocapsouza]

Hi domus123,

The models were trained on TPU v3-8 using TF 1.14 and the code from original BERT repository: https://github.com/google-research/bert/blob/master/run_pretraining.py.

Unfortunately I'm not a TF user so I can really help you with this problem. Did you manage to solve it?

[israellaurensi]

Any update on this? I'm having the same issue trying to convert the metadata to .pb

[domus123]

Hi, wasn't able to convert here. Seem's that tensorflow can convert GPU to CPU, but I don't have knowledge about TPU's