Open Zed-Wu opened 1 year ago
I find that in the unit test (e.g., transformer_network_test.py), the network_state is randomly sampled as:
network_state = tensor_spec.sample_spec_nest( network.state_spec, outer_dims=[BATCH_SIZE])
If I want to do inference in my own environment, what are the initial values of the network_state?
I find that in the unit test (e.g., transformer_network_test.py), the network_state is randomly sampled as:
If I want to do inference in my own environment, what are the initial values of the network_state?