diff --git a/mindspore/ccsrc/pipeline/jit/pipeline.cc b/mindspore/ccsrc/pipeline/jit/pipeline.cc index d0993f0fcc0..e3f383213b8 100644 --- a/mindspore/ccsrc/pipeline/jit/pipeline.cc +++ b/mindspore/ccsrc/pipeline/jit/pipeline.cc @@ -68,7 +68,7 @@ using mindspore::abstract::AbstractTuplePtr; const char IR_TYPE_ANF[] = "anf_ir"; const char IR_TYPE_ONNX[] = "onnx_ir"; -const char IR_TYPE_BINARY[] = "binary_ir"; +const char IR_TYPE_MINDIR[] = "mind_ir"; ExecutorPyPtr ExecutorPy::executor_ = nullptr; std::mutex ExecutorPy::instance_lock_; @@ -222,7 +222,7 @@ py::bytes ExecutorPy::GetFuncGraphProto(const std::string &phase, const std::str return proto_str; } - if (ir_type == IR_TYPE_BINARY) { + if (ir_type == IR_TYPE_MINDIR) { std::string proto_str = GetBinaryProtoString(fg_ptr); if (proto_str.empty()) { MS_LOG(EXCEPTION) << "Graph proto is empty."; diff --git a/mindspore/train/quant/quant.py b/mindspore/train/quant/quant.py index 45000d01948..a17d114550e 100644 --- a/mindspore/train/quant/quant.py +++ b/mindspore/train/quant/quant.py @@ -445,15 +445,17 @@ def export(network, *inputs, file_name, mean=127.5, std_dev=127.5, file_format=' file_name (str): File name of model to export. mean (int): Input data mean. Default: 127.5. std_dev (int, float): Input data variance. Default: 127.5. - file_format (str): MindSpore currently supports 'GEIR', 'ONNX' and 'BINARY' format for exported + file_format (str): MindSpore currently supports 'GEIR', 'ONNX' and 'MINDIR' format for exported quantization aware model. Default: 'GEIR'. - GEIR: Graph Engine Intermidiate Representation. An intermidiate representation format of Ascend model. - - BINARY: Binary format for model. An intermidiate representation format for models. + - MINDIR: MindSpore Native Intermidiate Representation for Anf. An intermidiate representation format + for MindSpore models. + Recommended suffix for output file is '.mindir'. """ supported_device = ["Ascend", "GPU"] - supported_formats = ['GEIR', 'BINARY'] + supported_formats = ['GEIR', 'MINDIR'] mean = validator.check_type("mean", mean, (int, float)) std_dev = validator.check_type("std_dev", std_dev, (int, float)) diff --git a/mindspore/train/serialization.py b/mindspore/train/serialization.py index e07cfa94c56..2783ac68697 100644 --- a/mindspore/train/serialization.py +++ b/mindspore/train/serialization.py @@ -453,17 +453,19 @@ def export(net, *inputs, file_name, file_format='GEIR'): net (Cell): MindSpore network. inputs (Tensor): Inputs of the `net`. file_name (str): File name of model to export. - file_format (str): MindSpore currently supports 'GEIR', 'ONNX' and 'BINARY' format for exported model. + file_format (str): MindSpore currently supports 'GEIR', 'ONNX' and 'MINDIR' format for exported model. - GEIR: Graph Engine Intermidiate Representation. An intermidiate representation format of Ascend model. - ONNX: Open Neural Network eXchange. An open format built to represent machine learning models. - - BINARY: Binary format for model. An intermidiate representation format for models. + - MINDIR: MindSpore Native Intermidiate Representation for Anf. An intermidiate representation format + for MindSpore models. + Recommended suffix for output file is '.mindir'. """ logger.info("exporting model file:%s format:%s.", file_name, file_format) check_input_data(*inputs, data_class=Tensor) - supported_formats = ['GEIR', 'ONNX', 'BINARY'] + supported_formats = ['GEIR', 'ONNX', 'MINDIR'] if file_format not in supported_formats: raise ValueError(f'Illegal file format {file_format}, it must be one of {supported_formats}') # switch network mode to infer when it is training @@ -485,10 +487,10 @@ def export(net, *inputs, file_name, file_format='GEIR'): with open(file_name, 'wb') as f: os.chmod(file_name, stat.S_IWUSR | stat.S_IRUSR) f.write(onnx_stream) - elif file_format == 'BINARY': # file_format is 'BINARY' - phase_name = 'export.binary' + elif file_format == 'MINDIR': # file_format is 'MINDIR' + phase_name = 'export.mindir' graph_id, _ = _executor.compile(net, *inputs, phase=phase_name, do_convert=False) - onnx_stream = _executor._get_func_graph_proto(graph_id, 'binary_ir') + onnx_stream = _executor._get_func_graph_proto(graph_id, 'mind_ir') with open(file_name, 'wb') as f: os.chmod(file_name, stat.S_IWUSR | stat.S_IRUSR) f.write(onnx_stream) diff --git a/serving/example/export_model/add_model.py b/serving/example/export_model/add_model.py index caf354cc5a5..76b247062ee 100644 --- a/serving/example/export_model/add_model.py +++ b/serving/example/export_model/add_model.py @@ -36,7 +36,7 @@ y = np.ones(4).astype(np.float32) def export_net(): add = Net() output = add(Tensor(x), Tensor(y)) - export(add, Tensor(x), Tensor(y), file_name='tensor_add.pb', file_format='BINARY') + export(add, Tensor(x), Tensor(y), file_name='tensor_add.pb', file_format='MINDIR') print(x) print(y) print(output.asnumpy()) diff --git a/tests/st/serving/generate_model.py b/tests/st/serving/generate_model.py index f7d47392f6c..ec05514405e 100644 --- a/tests/st/serving/generate_model.py +++ b/tests/st/serving/generate_model.py @@ -62,14 +62,14 @@ def export_add_model(): net = AddNet() x = np.ones(4).astype(np.float32) y = np.ones(4).astype(np.float32) - export(net, Tensor(x), Tensor(y), file_name='add.pb', file_format='BINARY') + export(net, Tensor(x), Tensor(y), file_name='add.pb', file_format='MINDIR') def export_bert_model(): net = BertModel(bert_net_cfg, False) input_ids = np.random.randint(0, 1000, size=(2, 32), dtype=np.int32) segment_ids = np.zeros((2, 32), dtype=np.int32) input_mask = np.zeros((2, 32), dtype=np.int32) - export(net, Tensor(input_ids), Tensor(segment_ids), Tensor(input_mask), file_name='bert.pb', file_format='BINARY') + export(net, Tensor(input_ids), Tensor(segment_ids), Tensor(input_mask), file_name='bert.pb', file_format='MINDIR') if __name__ == '__main__': export_add_model() diff --git a/tests/ut/python/utils/test_serialize.py b/tests/ut/python/utils/test_serialize.py index 7f85695a194..d0b19cb0c4b 100644 --- a/tests/ut/python/utils/test_serialize.py +++ b/tests/ut/python/utils/test_serialize.py @@ -322,10 +322,10 @@ def test_export(): @non_graph_engine -def test_binary_export(): +def test_mindir_export(): net = MYNET() input_data = Tensor(np.random.randint(0, 255, [1, 3, 224, 224]).astype(np.float32)) - export(net, input_data, file_name="./me_binary_export.pb", file_format="BINARY") + export(net, input_data, file_name="./me_binary_export.mindir", file_format="MINDIR") class PrintNet(nn.Cell):