creating a model with concat_pool = True creates a model that can't run on multiple tpu cores

[butchland]

Setting the concat_pool = True in cnn_learner (which is the default) creates a model that doesn't run on multiple tpu cores.

Setting it to concat_pool = False works fine however.

Based on the code, concat_pool=True uses fastai's AdaptiveConcatPool2D while setting it to False uses nn.AdaptiveAvgPool2D so it might be that using the fastai version causes a lowering or something that causes the model to not run at all.

Workaround: set the concat_pool=False when creating the custom head for a Learner for transfer learning (e.g. cnn_learner)