cyfwry

A little question about network trainning

3

Hello,I can not understand one sentence in your paper,'When the training error keeps unchanged in five sequential epochs, we merge the parameters of each batch normalization into the adjacent convolution...

bugfix: qk_scale is not used by swin FMHA.

2

Fixed issue #322. Now qk_scale can be used by swin FMHA. Signed-off-by: liangtao07

### Description ```shell branch: v5.0 gpu: T4 ``` ### Reproduced Steps ```shell 1. clone and compile FasterTransformer 2. cd examples/pytorch/swin 3. modify QK_SCALE in Swin-Transformer-Quantization/SwinTransformer/configs/swin_tiny_patch4_window7_224.yaml, e.g. QK_SCALE=2.0 4. sh run_test.sh...