New benchmark numbers

This commit is contained in:
James Betker 2021-02-08 08:09:41 -07:00
parent 784b96c059
commit 39fd755baa

View File

@ -231,8 +231,15 @@ def register_vqvae3_hard_switch(opt_net, opt):
def performance_test():
# For breadth=32:
# Custom_cuda_naive: 28.9s
# Torch_native: 29.2s
#
# For breadth=8
# Custom_cuda_naive: 18.4s
# Torch_native: 10s
cfg = {
'mode': 'lambda2',
'mode': 'lambda',
'breadth': 8,
'hard_enabled': True,
'dropout': 0.4