Closed ZhenYangIACAS closed 4 years ago
Hi thanks for asking: -)
Their scripts are not the same (--init-type adaptive
is changed to --init-type adaptive-profiling
for profiling).
The limitation is set at:
https://github.com/LiyuanLucasLiu/Transformer-Clinic/blob/a302132c52084d730aca5822abb58228a04b6bb8/fairseq/fairseq/tasks/fairseq_task.py#L265
Hello, I want to figure out how do you make sure that only performing one step forward pass in profiling phase in your code? Since the running scripts for profiling and training are the same, I cannot find where do you set the limitation for profiling.