wehub-resource-sync
151 lines
5.1 KiB
Python
151 lines
5.1 KiB
Python
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# 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.
|
|
|
|
from __future__ import annotations
|
|
|
|
import os
|
|
import unittest
|
|
from functools import partial
|
|
|
|
import numpy as np
|
|
from program_config import ProgramConfig, TensorConfig
|
|
from trt_layer_auto_scan_test import TrtLayerAutoScanTest
|
|
|
|
import paddle.inference as paddle_infer
|
|
|
|
|
|
class TrtConvertBmmTest_dynamic(TrtLayerAutoScanTest):
|
|
def sample_program_configs(self):
|
|
def generate_input(shape):
|
|
return np.random.random(shape).astype(np.float32)
|
|
|
|
for batch in [10, 11, 12, 13, 14, 15]:
|
|
for trans_x in [False]:
|
|
for trans_y in [False]:
|
|
input1_shape = [batch, 350, 75]
|
|
input2_shape = [batch, 75, 25]
|
|
dics = [{}]
|
|
ops_config = [
|
|
{
|
|
"op_type": "bmm",
|
|
"op_inputs": {
|
|
"X": ["input1_data"],
|
|
"Y": ["input2_data"],
|
|
},
|
|
"op_outputs": {"Out": ["output_data"]},
|
|
"op_attrs": dics[0],
|
|
}
|
|
]
|
|
ops = self.generate_op_config(ops_config)
|
|
|
|
program_config = ProgramConfig(
|
|
ops=ops,
|
|
weights={},
|
|
inputs={
|
|
"input2_data": TensorConfig(
|
|
data_gen=partial(generate_input, input2_shape)
|
|
),
|
|
"input1_data": TensorConfig(
|
|
data_gen=partial(generate_input, input1_shape)
|
|
),
|
|
},
|
|
outputs=["output_data"],
|
|
)
|
|
|
|
yield program_config
|
|
|
|
def generate_dynamic_shape(self, attrs):
|
|
self.dynamic_shape.min_input_shape = {
|
|
"input1_data": [10, 350, 75],
|
|
"input2_data": [10, 75, 25],
|
|
}
|
|
self.dynamic_shape.max_input_shape = {
|
|
"input1_data": [100, 350, 75],
|
|
"input2_data": [100, 75, 25],
|
|
}
|
|
self.dynamic_shape.opt_input_shape = {
|
|
"input1_data": [15, 350, 75],
|
|
"input2_data": [15, 75, 25],
|
|
}
|
|
return self.dynamic_shape
|
|
|
|
def sample_predictor_configs(
|
|
self, program_config, run_pir=False
|
|
) -> tuple[paddle_infer.Config, list[int], float]:
|
|
def clear_dynamic_shape():
|
|
self.dynamic_shape.min_input_shape = {}
|
|
self.dynamic_shape.max_input_shape = {}
|
|
self.dynamic_shape.opt_input_shape = {}
|
|
|
|
def generate_trt_nodes_num(attrs, dynamic_shape):
|
|
if dynamic_shape:
|
|
return 1, 3
|
|
else:
|
|
return 0, 4
|
|
|
|
attrs = [
|
|
program_config.ops[i].attrs for i in range(len(program_config.ops))
|
|
]
|
|
|
|
clear_dynamic_shape()
|
|
if not run_pir:
|
|
self.trt_param.precision = paddle_infer.PrecisionType.Float32
|
|
program_config.set_input_type(np.float32)
|
|
yield (
|
|
self.create_inference_config(),
|
|
generate_trt_nodes_num(attrs, False),
|
|
1e-5,
|
|
)
|
|
self.trt_param.precision = paddle_infer.PrecisionType.Half
|
|
program_config.set_input_type(np.float16)
|
|
yield (
|
|
self.create_inference_config(),
|
|
generate_trt_nodes_num(attrs, False),
|
|
(1e-2, 1e-2),
|
|
)
|
|
|
|
# The output has little diff between gpu and trt in CI-Windows-Inference
|
|
tol_fp32 = 1e-4
|
|
tol_half = 1e-2
|
|
if os.name == 'nt':
|
|
tol_fp32 = 1e-2
|
|
tol_half = 1e-2
|
|
# for dynamic_shape
|
|
self.generate_dynamic_shape(attrs)
|
|
self.trt_param.precision = paddle_infer.PrecisionType.Float32
|
|
program_config.set_input_type(np.float32)
|
|
yield (
|
|
self.create_inference_config(),
|
|
generate_trt_nodes_num(attrs, True),
|
|
tol_fp32,
|
|
)
|
|
self.trt_param.precision = paddle_infer.PrecisionType.Half
|
|
program_config.set_input_type(np.float16)
|
|
yield (
|
|
self.create_inference_config(),
|
|
generate_trt_nodes_num(attrs, True),
|
|
(tol_half, tol_half),
|
|
)
|
|
|
|
def add_skip_trt_case(self):
|
|
pass
|
|
|
|
def test(self):
|
|
self.add_skip_trt_case()
|
|
self.run_test(run_pir=True)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|