[Question] After updating to 0.5.1 multi-gpu support have not been working

Hi! Thank you for the great lib!

After updating from 0.2.4 to 0.5.1 only one gpu out of two is used. The code almost the same:

hyper_parameters = BLSTMModel.get_default_hyper_parameters()
hyper_parameters['layer_bi_lstm']['units'] = 1024

model = BLSTMModel(embedding, hyper_parameters=hyper_parameters)
model.build_model(X_train, y_train)
model.build_multi_gpu_model(gpus=2, x_train=X_train, y_train=y_train, x_validate=X_valid, y_validate=y_valid)
model.fit(X_train, y_train, epochs=15, batch_size=512, x_validate=X_valid, y_validate=y_valid, callbacks=[tf_board_callback, checkpoint_callback])

NVtop and nvidia-smi showing that only one gpu is working. In the previous version I have both of my gpus been used.

[jeshuren]

I am also facing the same issue. Would be great if someone could help.

UPDATE:

Try without calling the build_model() method. It works!

model = BLSTMModel(embedding, hyper_parameters=hyper_parameters)

model.build_multi_gpu_model(gpus=2, x_train=X_train, y_train=y_train, x_validate=X_valid, y_validate=y_valid)

model.fit(X_train, y_train, epochs=15, batch_size=512, x_validate=X_valid, y_validate=y_valid, callbacks=[tf_board_callback, checkpoint_callback])

[sld]

Thanks! This https://github.com/BrikerMan/Kashgari/issues/170#issuecomment-513875786 worked!