kyoguan

bug in the gcc lib

2

We found the gcc lib has a bug, call the system call futex without the FUTEX_PRIVATE_FLAG flag. This would block the thread after mesh, because the phys address changed. clang'...

have you tryed training with the full dataset?

10

this result is training on single gpu. this result is training on 8 gpu. (BTW: you need to set BN = None, or you would got a strange result ,...

single GPU is better?

3

batch_size = 32*6 and num_gpus = 6(train_multi_gpus.py) batch_size = 32 (train.py) it seems the single gpu is mush faster and the training curve is smoother.