Skip to content
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.

Sign up for GitHub

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

Jump to bottom

GNA MTL mixed precision #13166

Merged
merged 5 commits into from
Nov 8, 2022
Merged

GNA MTL mixed precision #13166

merged 5 commits into from
Nov 8, 2022

Conversation

sirzabek
Copy link
Contributor

@sirzabek sirzabek commented Sep 22, 2022 •
edited
Loading

Details:

  • enable int8 weights for convolutions with int16 inputs on MTL platform.

Tickets:

  • 93967

@sirzabek sirzabek requested review from a team as code owners September 22, 2022 10:27
@sirzabek sirzabek removed request for a team September 22, 2022 10:29
@sirzabek sirzabek added WIP work in progress do_not_review labels Oct 13, 2022
@sirzabek sirzabek changed the title Gna mtl low precision GNA MTL mixed precision Oct 13, 2022
@sirzabek sirzabek force-pushed the gna_mtl_low_precision branch 6 times, most recently from f99084a to 4d70291 Compare October 24, 2022 13:39
@sirzabek sirzabek force-pushed the gna_mtl_low_precision branch 4 times, most recently from 0aa947d to ec30437 Compare November 2, 2022 08:03
@sirzabek sirzabek force-pushed the gna_mtl_low_precision branch 2 times, most recently from cc06ef9 to 4033a7d Compare November 3, 2022 11:47
marcinkusm
marcinkusm reviewed Nov 4, 2022
View reviewed changes
Copy link
Contributor

@marcinkusm marcinkusm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

please update at least tests and unitialized variables

src/plugins/intel_gna/src/frontend/layer_quantizer.hpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/frontend/layer_quantizer.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/frontend/layer_quantizer.cpp Outdated Show resolved Hide resolved
src/tests_deprecated/unit/engines/gna/i16_quantisation_test.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/frontend/layer_quantizer.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/frontend/weights_converter.cpp Show resolved Hide resolved
src/plugins/intel_gna/src/frontend/weights_converter.hpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/gna_plugin.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/gna_plugin.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/frontend/weights_converter.cpp Outdated Show resolved Hide resolved
@sirzabek sirzabek removed WIP work in progress do_not_review labels Nov 7, 2022
kbruniec
kbruniec approved these changes Nov 7, 2022
View reviewed changes
Copy link
Contributor

@kbruniec kbruniec left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

src/plugins/intel_gna/src/frontend/layer_quantizer.hpp Outdated Show resolved Hide resolved
src/tests_deprecated/unit/engines/gna/graph_tools/graph_copy_tests.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/tests/unit/gna_hw_precision_test.cpp
@@ -26,6 +26,7 @@ class GNAPluginForPrecisionTest : public GNAPluginNS::GNAPlugin {
}
void set_low_precision_input() {
this->gnaFlags->input_low_precision = true;
this->config.gnaFlags.input_low_precision = true;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

we should clean it up in future (get rid of one of this two copies).

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I agree, but we need to sort it out in the plugin, not in the test.

.../intel_gna/tests/functional/shared_tests_instances/single_layer_tests/conv_low_precision.cpp Outdated Show resolved Hide resolved
src/plugins/intel_gna/src/layers/gna_fake_quantize_layer.hpp Outdated Show resolved Hide resolved
@sirzabek sirzabek requested a review from a team November 7, 2022 11:49
marcinkusm
marcinkusm approved these changes Nov 7, 2022
View reviewed changes
Copy link
Contributor

@marcinkusm marcinkusm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Few minor changes needed, would be better to have few smaller PR-s.

@sirzabek sirzabek merged commit 3ea77b9 into openvinotoolkit:master Nov 8, 2022
tadamowicz pushed a commit to tadamowicz/openvino that referenced this pull request Aug 30, 2023
@sirzabek @marcinkacprzak @dood-apo
GNA MTL mixed precision (openvinotoolkit#13166)
aeb8e00 
* [GNA] Enable int8 weights for convolutions

* [GNA] Implement convolution low precision tests for MTL

Co-authored-by: Marcin Kacprzak <marcin.kacprzak@intel.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@marcinkacprzak marcinkacprzak marcinkacprzak left review comments

@kbruniec kbruniec kbruniec approved these changes

@marcinkusm marcinkusm marcinkusm approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
4 participants
@sirzabek @kbruniec @marcinkusm @marcinkacprzak