File size: 3,037 Bytes
5000658 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 |
# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import onnx
import tensorrt as trt
from onnx import TensorProto, helper
def trt_dtype_to_onnx(dtype):
if dtype == trt.float16:
return TensorProto.DataType.FLOAT16
if dtype == trt.bfloat16:
return TensorProto.DataType.BFLOAT16
elif dtype == trt.float32:
return TensorProto.DataType.FLOAT
elif dtype == trt.int32:
return TensorProto.DataType.INT32
elif dtype == trt.int64:
return TensorProto.DataType.INT64
elif dtype == trt.bool:
return TensorProto.DataType.BOOL
else:
raise TypeError("%s is not supported" % dtype)
def to_onnx(network, path, name: str = None):
if name is None:
name = "debug_graph"
inputs = []
for i in range(network.num_inputs):
network_input = network.get_input(i)
inputs.append(
helper.make_tensor_value_info(
network_input.name, trt_dtype_to_onnx(network_input.dtype),
list(network_input.shape)))
outputs = []
for i in range(network.num_outputs):
network_output = network.get_output(i)
outputs.append(
helper.make_tensor_value_info(
network_output.name, trt_dtype_to_onnx(network_output.dtype),
list(network_output.shape)))
nodes = []
for i in range(network.num_layers):
layer = network.get_layer(i)
layer_inputs = []
for j in range(layer.num_inputs):
ipt = layer.get_input(j)
if ipt is not None:
layer_inputs.append(layer.get_input(j).name)
layer_outputs = [
layer.get_output(j).name for j in range(layer.num_outputs)
]
nodes.append(
helper.make_node(str(layer.type),
name=layer.name,
inputs=layer_inputs,
outputs=layer_outputs,
domain="com.nvidia"))
onnx_model = helper.make_model(helper.make_graph(nodes,
name,
inputs,
outputs,
initializer=None),
producer_name='NVIDIA')
onnx.save(onnx_model, path)
|