合并请求 · PaddlePaddle / Paddle

Add bn and relu fuse pass

!22048 · opened 1月 02, 2020 by saxon_zh AMP
- 已合并
- 12
更新于2月 10, 2020
Support fp16 in GPU impl of fused_elemwise_activation_op.

!20636 · opened 10月 15, 2019 by saxon_zh AMP
- 已合并
- 0
更新于11月 14, 2019
Add NHWC datalayout support on conv2d without data transformation.

!20597 · opened 10月 14, 2019 by saxon_zh AMP
- 已合并
- 12
更新于2月 10, 2020
Add interface so user can get scaled loss when they use customized loss.

!20571 · opened 10月 12, 2019 by saxon_zh AMP
- 已合并
- 0
更新于10月 15, 2019
Add dtype for coalesce_tensor_op

!20016 · opened 9月 26, 2019 by saxon_zh AMP
- 已合并
- 0
更新于9月 26, 2019
Expose `mutable_data` as python binding

!19932 · opened 9月 23, 2019 by saxon_zh AMP
- 已合并
- 1
更新于9月 27, 2019
Add fp16 support for pad and split

!19881 · opened 9月 19, 2019 by saxon_zh AMP
- 已合并
- 0
更新于2月 05, 2021
Optimize amp for multi-gpu to enable FP16 gradients transfer across gpus

!19714 · opened 9月 09, 2019 by saxon_zh AMP
- 已合并
- 8
更新于9月 19, 2019
Fix float16 optimizer.

!19682 · opened 9月 06, 2019 by saxon_zh AMP
- 已合并
- 0
更新于9月 17, 2019
Change backward_guard to optimize_guard to maximize the allreduce overlap.

!19506 · opened 8月 28, 2019 by saxon_zh AMP
- 已合并
- 3
更新于9月 17, 2019
init new amp

!18596 · opened 7月 11, 2019 by saxon_zh AMP
- 已合并
- 1
更新于9月 17, 2019
Register fp16 for concat_op

!18563 · opened 7月 09, 2019 by saxon_zh AMP
- 已合并
- 1
更新于9月 17, 2019
custom black white list in AMP

!18377 · opened 6月 27, 2019 by saxon_zh AMP NVIDIA
- 已合并
- 0
更新于9月 17, 2019
init auto loss scaling

!17194 · opened 5月 05, 2019 by saxon_zh AMP
- 已合并
- 4
更新于9月 17, 2019
Init mixed precision training interface

!16856 · opened 4月 15, 2019 by saxon_zh AMP
- 已合并
- 34
更新于2月 10, 2020
Impl fp16 compute kernel for slice_op

!16206 · opened 3月 14, 2019 by saxon_zh AMP
- 已合并
- 0
更新于2月 10, 2020
Fp16 training

!14992 · opened 12月 21, 2018 by saxon_zh AMP
- 已合并
- 12
更新于2月 10, 2020
[Feature] support mix precision training for resnet

!14899 · opened 12月 14, 2018 by saxon_zh AMP
- 已合并
- 39
更新于2月 10, 2020
Add half precision float16 data type

!5716 · opened 11月 17, 2017 by saxon_zh AMP
- 已合并
- 33
更新于2月 10, 2020

PaddlePaddle / Paddle 大约 1 年 前同步成功

PaddlePaddle / Paddle
大约 1 年前同步成功