Specify precision for the openvino core object

The latest processors like sapphire rapids assume default
precision to be BF16. Hence, the accuracy for F32 models with
default config is impacted.

Adding changes to specify the precision, to avoid accuracy
issues across processors.

Tracked-On: OAM-111649
Signed-off-by: Anoob Anto K <[email protected]>

IENetwork.cpp

@@ -39,6 +39,7 @@ bool IENetwork::createNetwork(std::shared_ptr<ov::Model> network, const std::str
         ALOGE("Invalid Network pointer");
         return false;
     } else {
+        ie.set_property(deviceStr, {{ov::hint::inference_precision.name(), "f32"}});
         ov::CompiledModel compiled_model = ie.compile_model(network, deviceStr);
         ALOGD("createNetwork is done....");
 #if __ANDROID__
@@ -76,6 +77,7 @@ void IENetwork::loadNetwork(const std::string& modelName) {
 
     ALOGD("loading infer request for Intel Device Type : %s", deviceStr.c_str());
 
+    ie.set_property(deviceStr, {{ov::hint::inference_precision.name(), "f32"}});
     ov::CompiledModel compiled_model = ie.compile_model(modelName, deviceStr);
     mInferRequest = compiled_model.create_infer_request();
     isLoaded = true;