hongzimao/decima-sim

A question about actor_network

Opened this issue · 2 comments

When I use a large dataset to run decima, I get an error "BiasGrad requires tensor size <= int32 max", and I wonder if you met this error? As the data set increases, so does the size of the neural network. Do you have any idea to solve this error?

I didn't see this error before but it looks like you hit an Inf somewhere. Could you narrow down to where the error is emitted from?

When the program is computing, it will crash due to excessive memory consumption, so I made some changes such as decreasing the interval of gradient calculating and reducing the size of networks. And when decima makes too much schedule decisions before calculating gradient, it will crash for this error when calculating gradient.