Skip to content

Commit 065e4ce

Browse files
Merge pull request #237 from yuanzhedong/yz/dev/mrcnn_new_rule
[MRCNN] allow lr 0.01 for when gbs < 16
2 parents 59d2b6d + f89c380 commit 065e4ce

File tree

1 file changed

+4
-2
lines changed

1 file changed

+4
-2
lines changed

mlperf_logging/compliance_checker/training_2.0.0/closed_maskrcnn.yaml

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,8 @@
11
- BEGIN:
22
CODE: >
33
s.update({
4-
'initialized_tensors': []
4+
'initialized_tensors': [],
5+
'global_batch_size': None,
56
})
67
- KEY:
78
NAME: weights_initialization
@@ -40,11 +41,12 @@
4041
NAME: global_batch_size
4142
REQ: EXACTLY_ONE
4243
CHECK: " v['value'] > 0"
44+
POST: " s['global_batch_size'] = v['value'] "
4345

4446
- KEY:
4547
NAME: opt_base_learning_rate
4648
REQ: EXACTLY_ONE
47-
CHECK: " is_integer(v['value'] / 0.02) "
49+
CHECK: " is_integer(v['value'] / 0.02) or ( s['global_batch_size'] < 16 and is_integer(0.02 / v['value']) )"
4850

4951
- KEY:
5052
NAME: opt_learning_rate_warmup_steps

0 commit comments

Comments
 (0)