Spaces:
Running
on
L4
Running
on
L4
File size: 3,143 Bytes
5b4c852 |
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 82 83 84 85 86 87 88 89 90 91 92 93 |
# Copyright (c) 2020 Mobvoi Inc (Di Wu)
# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu)
#
# 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 os
import argparse
import glob
import yaml
import torch
def get_args():
parser = argparse.ArgumentParser(description='average model')
parser.add_argument('--dst_model', required=True, help='averaged model')
parser.add_argument('--src_path',
required=True,
help='src model path for average')
parser.add_argument('--val_best',
action="store_true",
help='averaged model')
parser.add_argument('--num',
default=5,
type=int,
help='nums for averaged model')
args = parser.parse_args()
print(args)
return args
def main():
args = get_args()
val_scores = []
if args.val_best:
yamls = glob.glob('{}/*.yaml'.format(args.src_path))
yamls = [
f for f in yamls
if not (os.path.basename(f).startswith('train')
or os.path.basename(f).startswith('init'))
]
for y in yamls:
with open(y, 'r') as f:
dic_yaml = yaml.load(f, Loader=yaml.BaseLoader)
loss = float(dic_yaml['loss_dict']['loss'])
epoch = int(dic_yaml['epoch'])
step = int(dic_yaml['step'])
tag = dic_yaml['tag']
val_scores += [[epoch, step, loss, tag]]
sorted_val_scores = sorted(val_scores,
key=lambda x: x[2],
reverse=False)
print("best val (epoch, step, loss, tag) = " +
str(sorted_val_scores[:args.num]))
path_list = [
args.src_path + '/epoch_{}_whole.pt'.format(score[0])
for score in sorted_val_scores[:args.num]
]
print(path_list)
avg = {}
num = args.num
assert num == len(path_list)
for path in path_list:
print('Processing {}'.format(path))
states = torch.load(path, map_location=torch.device('cpu'))
for k in states.keys():
if k not in avg.keys():
avg[k] = states[k].clone()
else:
avg[k] += states[k]
# average
for k in avg.keys():
if avg[k] is not None:
# pytorch 1.6 use true_divide instead of /=
avg[k] = torch.true_divide(avg[k], num)
print('Saving to {}'.format(args.dst_model))
torch.save(avg, args.dst_model)
if __name__ == '__main__':
main()
|