CUDA ERROR, run out of memory.

[alexpm94]

@watersink I tested training on my other system with 30GB ram and Quadro M4000 graphics card and it worked there.
Thanks!

Originally posted by @ghost in #1 (comment)

I tried it to train using GTX970M 16GB RAM and it failed. I tried Google Colab and it failed too. I also tried using half of the dataset, setting per_process_gpu_memory_fraction=0.7 but it did not work. There is no other process using GPU. I am looking at the code trying to figure out is there something defining the input shape but it seems ok. Have you ever encountered the same problem?

Thanks in advance.