-
Notifications
You must be signed in to change notification settings - Fork 240
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Divide as a * (1.0 / b)
during weight compression
#3055
base: develop
Are you sure you want to change the base?
Divide as a * (1.0 / b)
during weight compression
#3055
Conversation
a * (1.0 / b)
during weight compression
c12cd32
to
70ec840
Compare
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
nncf/quantization/algorithms/weight_compression/weight_lowering.py
Outdated
Show resolved
Hide resolved
nncf/quantization/algorithms/weight_compression/scale_estimation.py
Outdated
Show resolved
Hide resolved
The problem with division is resolved by aligning OV division with NP division by disabling |
Changes
fns.reciprocal(x) = 1.0 / x
operationa / b
witha * fns.reciprocal(b)
during weight compression -related computations.Reason for changes
Align with how division is performed in OpenVINO. Prerequisite to #2727. This should be merged first in order to later track only the differences introduced by the OV computations.
Related tickets
139047
Tests
Added a test comparing inverted division result to a reference value.
NNCF/job/manual/job/post_training_weight_compression/271/
NNCF/job/manual/job/post_training_quantization/576/
Also weekly validation is requested in 158012