* support fp16 training * Use compiled training program * Change timing ips. * Use dali * add pure fp16 training * fix a bug, which will not use fuse pass using pure fp16 training. * modify code as review * modify loss, so that it will use different loss when using pure fp16 training. * remove some fluid API * add static optimizer. |
||
---|---|---|
.. | ||
dali.py | ||
optimizer.py | ||
program.py | ||
run_dali.sh | ||
save_load.py | ||
train.py |