Memory efficient version that can handle up to ~25000 neurons

JavierGalvan9 / V1_GLIF_model

TensorFlow implementation of the new version from the V1 column from the Allen Institute

0 stars 4 forks source link

Memory efficient version that can handle up to ~25000 neurons #10

Closed shixnya closed 10 months ago

shixnya commented 10 months ago

I made a 'prepare_sparse_weights' method for the recurrent network and had it run before the model was executed. Note that this needs to be done within the gradient tape scope when it is trained.