diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-09-10 20:26:48 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-09-11 13:47:54 +0000 |
commit | 16cf3c2318978580006633ea4d5eccdd8ba1194a (patch) | |
tree | bf2614fcf8ab984f59f85d8502c65a57527ca1a6 /tests/SimpleTensorPrinter.h | |
parent | cffb2a34d89be3b175a8ba1ee8083b8ec256a633 (diff) | |
download | ComputeLibrary-16cf3c2318978580006633ea4d5eccdd8ba1194a.tar.gz |
COMPMID-3764: Failures on fp16 for L2NormalizationLayer when zero inputs
Alter the default lower bound used for the norm from 1e-12 to 1e-6 to be
representable by the half precision dynamic range.
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I8d3103b8345eb4c464a76b4f4ba5ef596d81da93
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3960
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'tests/SimpleTensorPrinter.h')
0 files changed, 0 insertions, 0 deletions