f86a797b10
* [kws] add static quantize * refine lint error in shuffle_list.py * refine lint * fix topo
135 lines
5.3 KiB
Python
135 lines
5.3 KiB
Python
# Copyright (c) 2021 Binbin Zhang(binbzha@qq.com)
|
|
#
|
|
# 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 print_function
|
|
|
|
import argparse
|
|
import copy
|
|
import logging
|
|
import os
|
|
import sys
|
|
|
|
import torch
|
|
import yaml
|
|
from torch.utils.data import DataLoader
|
|
|
|
from kws.dataset.dataset import Dataset
|
|
from kws.model.kws_model import init_model
|
|
from kws.utils.checkpoint import load_checkpoint
|
|
|
|
|
|
def get_args():
|
|
parser = argparse.ArgumentParser(description='recognize with your model')
|
|
parser.add_argument('--config', required=True, help='config file')
|
|
parser.add_argument('--test_data', required=True, help='test data file')
|
|
parser.add_argument('--checkpoint', required=True, help='checkpoint model')
|
|
parser.add_argument('--num_workers',
|
|
default=0,
|
|
type=int,
|
|
help='num of subprocess workers for reading')
|
|
parser.add_argument('--pin_memory',
|
|
action='store_true',
|
|
default=False,
|
|
help='Use pinned memory buffers used for reading')
|
|
parser.add_argument('--prefetch',
|
|
default=100,
|
|
type=int,
|
|
help='prefetch number')
|
|
parser.add_argument('--script_model',
|
|
required=True,
|
|
help='output script model')
|
|
args = parser.parse_args()
|
|
print(args)
|
|
return args
|
|
|
|
|
|
def main():
|
|
args = get_args()
|
|
logging.basicConfig(level=logging.DEBUG,
|
|
format='%(asctime)s %(levelname)s %(message)s')
|
|
os.environ['CUDA_VISIBLE_DEVICES'] = str("-1")
|
|
|
|
with open(args.config, 'r') as fin:
|
|
configs = yaml.load(fin, Loader=yaml.FullLoader)
|
|
|
|
test_conf = copy.deepcopy(configs['dataset_conf'])
|
|
test_conf['filter_conf']['max_length'] = 102400
|
|
test_conf['filter_conf']['min_length'] = 0
|
|
test_conf['speed_perturb'] = False
|
|
test_conf['spec_aug'] = False
|
|
test_conf['shuffle'] = False
|
|
test_conf['feature_extraction_conf']['dither'] = 0.0
|
|
test_conf['batch_conf']['batch_size'] = 1
|
|
|
|
test_dataset = Dataset(args.test_data, test_conf)
|
|
test_data_loader = DataLoader(test_dataset,
|
|
batch_size=None,
|
|
pin_memory=args.pin_memory,
|
|
num_workers=args.num_workers,
|
|
prefetch_factor=args.prefetch)
|
|
|
|
# Init asr model from configs
|
|
model_fp32 = init_model(configs['model'])
|
|
load_checkpoint(model_fp32, args.checkpoint)
|
|
# model must be set to eval mode for static quantization logic to work
|
|
model_fp32.eval()
|
|
|
|
# Fuse the activations to preceding layers, where applicable.
|
|
# This needs to be done manually depending on the model architecture.
|
|
# Common fusions include `conv + relu` and `conv + batchnorm + relu`
|
|
print('================ Float 32 ======================')
|
|
print(model_fp32)
|
|
print('================ Float 32(fused) ===============')
|
|
model_fp32.fuse_modules()
|
|
print(model_fp32)
|
|
|
|
# attach a global qconfig, which contains information about what kind
|
|
# of observers to attach. Use 'fbgemm' for server inference and
|
|
# 'qnnpack' for mobile inference. Other quantization configurations such
|
|
# as selecting symmetric or assymetric quantization and MinMax or L2Norm
|
|
# calibration techniques can be specified here.
|
|
model_fp32.qconfig = torch.quantization.get_default_qconfig('qnnpack')
|
|
|
|
# Prepare the model for static quantization. This inserts observers in
|
|
# the model that will observe activation tensors during calibration.
|
|
model_fp32_prepared = torch.quantization.prepare(model_fp32)
|
|
|
|
# calibrate the prepared model to determine quantization parameters for
|
|
# activations in a real world setting, the calibration would be done with
|
|
# a representative dataset
|
|
with torch.no_grad():
|
|
for batch_idx, batch in enumerate(test_data_loader):
|
|
keys, feats, target, lengths = batch
|
|
logits = model_fp32_prepared(feats)
|
|
if batch_idx % 100 == 0:
|
|
print('Progress utts {}'.format(batch_idx))
|
|
sys.stdout.flush()
|
|
|
|
# Convert the observed model to a quantized model. This does several things:
|
|
# quantizes the weights, computes and stores the scale and bias value to be
|
|
# used with each activation tensor, and replaces key operators with
|
|
# quantized implementations.
|
|
print('=================== int8 ======================')
|
|
model_int8 = torch.quantization.convert(model_fp32_prepared)
|
|
print(model_int8)
|
|
|
|
print('================ int8(script) ==================')
|
|
script_model = torch.jit.script(model_int8)
|
|
script_model.save(args.script_model)
|
|
print(script_model)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|