IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU] Fix FP16 to FP32 data types converting (#7003)

Browse Source
This commit is contained in:
Sergey Shlyapnikov 2021-08-13 11:43:14 +03:00 committed by GitHub
parent af8dc644a8
commit cc15581cd5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
inference-engine/src/cldnn_engine/cldnn_engine.cpp
View File

@ -388,11 +388,12 @@ InferenceEngine::CNNNetwork clDNNEngine::CloneAndTransformNetwork(const Inferenc
OV_ITT_SCOPED_TASK(itt::domains::CLDNNPlugin, "clDNNEngine::TransformNetwork::LPT");
using namespace ngraph::pass::low_precision;
ngraph::pass::Manager manager;
// Conversion to FP32 might be needed for quantized models that face any fp16 related issues (e.g. overflow) for non-quantized layers
// With this key users can work-around such issues
if (!config.enable_fp16_for_quantized_models) {
ngraph::pass::Manager manager;
manager.register_pass<ngraph::pass::ConvertPrecision>(precisions_array {{ ngraph::element::f16, ngraph::element::f32 }});
manager.run_passes(nGraphFunc);
}
auto supportedPrecisions = std::vector<OperationPrecisionRestriction>({