簡介【免費(fèi)下載鏈接】geGEGraph Engine是面向昇騰的圖編譯器和執(zhí)行器提供了計(jì)算圖優(yōu)化、多流并行、內(nèi)存復(fù)用和模型下沉等技術(shù)手段加速模型執(zhí)行效率減少模型內(nèi)存占用。 GE 提供對 PyTorch、TensorFlow 前端的友好接入能力并同時支持 onnx、pb 等主流模型格式的解析與編譯。項(xiàng)目地址: https://gitcode.com/cann/ge構(gòu)造Graph相關(guān)的類。需要包含的頭文件#include graph/graph.hPublic成員函數(shù)explicit Graph(const std::string name) explicit Graph(const char *name) Graph() ~Graph() graphStatus AddControlEdge(GNode src_node, GNode dst_node) graphStatus AddDataEdge(GNode src_node, const int32_t src_port_index, GNode dst_node, const int32_t dst_port_index) GNode AddNodeByOp(const Operator op) graphStatus AddOp(const ge::Operator op) graphStatus Graph::CopyFrom(const Graph src_graph) static GraphPtr ConstructFromInputs(const std::vectorOperator inputs, const AscendString name) graphStatus Dump(DumpFormat format, std::ostream o_stream) const graphStatus DumpToFile(DumpFormat format, const AscendString suffix) const graphStatus FindOpByName(const std::string name, ge::Operator op) const graphStatus FindOpByName(const char_t *name, ge::Operator op) const graphStatus FindOpByType(const std::string type, std::vectorge::Operator ops) const graphStatus FindOpByType(const char_t *type, std::vectorge::Operator ops) const std::vectorGNode GetAllNodes() const graphStatus GetAttr(const AscendString name, AttrValue attr_value) const graphStatus GetAttr(const AscendString name, AttrValue attr_value) const graphStatus GetAllOpName(std::vectorstd::string op_name) const graphStatus GetAllOpName(std::vectorAscendString names) const std::vectorGNode GetDirectNode() const const std::string Graph::GetName() const graphStatus GetName(AscendString name) const bool IsValid() const graphStatus LoadFromFile(const std::string file_name) graphStatus LoadFromFile(const char_t *file_name) graphStatus LoadFromMem(const GraphBuffer graph_buffer) graphStatus LoadFromMem(const uint8_t *data, const size_t len) graphStatus LoadFromSerializedModelArray(const void *serialized_model, size_t size) graphStatus RemoveEdge(GNode src_node, const int32_t src_port_index, GNode dst_node, const int32_t dst_port_index) graphStatus RemoveNode(GNode node) graphStatus RemoveNode(GNode node, bool contain_subgraph) graphStatus SaveToFile(const std::string file_name) const graphStatus SaveToFile(const char_t *file_name) const graphStatus SaveToMem(GraphBuffer graph_buffer) const graphStatus SetAttr(const AscendString name, const AttrValue attr_value) Graph SetInputs(const std::vectorOperator inputs) void SetNeedIteration(bool need_iteration) Graph SetOutputs(const std::vectorOperator outputs) Graph SetOutputs(const std::vectorstd::pairOperator, std::vectorsize_t output_indexs) Graph SetOutputs(const std::vectorstd::pairge::Operator, std::string outputs) Graph SetOutputs(const std::vectorstd::pairge::Operator, AscendString outputs) Graph SetTargets(const std::vectorOperator targets)【免費(fèi)下載鏈接】geGEGraph Engine是面向昇騰的圖編譯器和執(zhí)行器提供了計(jì)算圖優(yōu)化、多流并行、內(nèi)存復(fù)用和模型下沉等技術(shù)手段加速模型執(zhí)行效率減少模型內(nèi)存占用。 GE 提供對 PyTorch、TensorFlow 前端的友好接入能力并同時支持 onnx、pb 等主流模型格式的解析與編譯。項(xiàng)目地址: https://gitcode.com/cann/ge創(chuàng)作聲明：本文部分內(nèi)容由AI輔助生成（AIGC），僅供參考