saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-29 23:06:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

epsilon attribute for layernormalization fusion (#2639)

Browse source
This commit is contained in:
liuziyue 2019-12-12 10:58:02 -08:00 committed by Tianlei Wu
parent d6f33dceb1
commit c7cd336917
1 changed files with 9 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
onnxruntime/core/optimizer/layer_norm_fusion.cc
View file

@ -153,7 +153,6 @@ Status LayerNormFusion::ApplyImpl(Graph& graph, bool& modified, int graph_level,
continue;
}
nodes_to_remove.push_back(add2_node);
// Traceback the add node to find reduceMean --> add
const Node* p_reduce_mean2 = nullptr;
@ -255,6 +254,15 @@ Status LayerNormFusion::ApplyImpl(Graph& graph, bool& modified, int graph_level,
layer_norm_input_defs,
{}, {}, kOnnxDomain);
// Get constant "epsilon" from "Add2" node if available. Else, default value will be used.
const ONNX_NAMESPACE::TensorProto* tensor_proto = graph_utils::GetConstantInitializer(graph, add2_node.MutableInputDefs()[1]->Name());
if (tensor_proto != nullptr) {
if (tensor_proto->data_type() == ONNX_NAMESPACE::TensorProto_DataType_FLOAT) {
const float* val = onnxruntime::make_unique<Initializer>(*tensor_proto)->data<float>();
layer_norm_node.AddAttribute("epsilon", val[0]);
}
}
// Assign provider to this new node. Provider should be same as the provider for old node.
layer_norm_node.SetExecutionProviderType(reduce_mean_node.GetExecutionProviderType());