Issues
- 3
The same batch size, different micro batches, the algorithm effects are inconsistent.
#35 opened by Kurama622 - 1
- 0
CUDA issues
#33 opened by linghan1997 - 2
RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same
#32 opened by joelrorseth - 4
Whether to support multi-machine training ?
#7 opened by yangpc615 - 1
[Question] Inference time speed up or not?
#31 opened by tuanmanh1410 - 0
- 4
Could torchgpipe run on CPU-only machines?
#29 opened by HuYang719 - 3
About the debugger for running torchgpipe
#26 opened by Real-ZeminJiang - 1
Issues on torchgpipe project and paper
#28 opened by xshaun - 0
Failed when trying to use auto partition on running resnet101-speed benchmark
#27 opened by GeorgeQ-Q - 6
- 1
- 7
- 1
torchpipe + megatron
#23 opened by eric-haibin-lin - 5
Work with DistributedDataParallel
#16 opened by YHRen - 0
Can torchgpipe work with Horovod?
#22 opened by RainFrost1 - 1
A more memory-efficient implementation
#19 opened by anxuthu - 3
Forward in eval() mode.
#17 opened by anxuthu - 2
About speedup
#18 opened by anxuthu - 6
Why does `Copy` compute gradients in reversed order
#15 opened by MlWoo - 2
How did you handle batch norm?
#14 opened by nirandaperera - 10
Using GPipe for Hessian Computation
#13 opened by singhsarvagya - 11
Checkpoint Issues
#12 opened by vibhatha - 5
Benchmark Performance for Baseline vs Pipeline-1
#11 opened by vibhatha - 11
Gpipe Benchmark
#10 opened by vibhatha - 6
convergence problem
#9 opened by yangpc615 - 8
KeyError in Stash
#8 opened by wentaozhu - 2
- 3
Reproducing GPipe accuracy results
#1 opened by 0xsamgreen - 1
Typo in documentation
#4 opened by Kyeongpil - 5
Balance Module
#3 opened by 842974287 - 4
Question about worker thread in GPipe
#2 opened by 842974287