AI/wekws
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

learning rate won't initiate from 0.001 when continuing training from checkpoint (#67)

Browse Source 
* add .gitattributes

* add long wav

* fix some bugs

* updated lint error

* back the hi_xiaowen/run.sh to the same

* remove the space

* better one

* remove 'num_keyword' parameter

* remove files

* flask8 examine

* override the score and compute_det file

* remove defaultdict

* remove import defaultdict

* learning rate won't initial from 0.001 when continuing training from checkpoint

* fix intent bug with initial learning rate != 0.001
This commit is contained in:
Cyan 2022-04-14 16:02:18 +08:00 committed by GitHub
parent 7d142b9528
commit 015748b94e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
kws/bin/train.py
View File

@ -157,7 +157,9 @@ def main():
infos = {}
start_epoch = infos.get('epoch', -1) + 1
cv_loss = infos.get('cv_loss', 0.0)
# get the last epoch lr
lr_last_epoch = infos.get('lr', configs['optim_conf']['lr'])
configs['optim_conf']['lr'] = lr_last_epoch
model_dir = args.model_dir
writer = None
if rank == 0: