Hugging Face's logo

Spaces:
ASLP-lab
/
OSUM-EChat
Running on Zero

App Files Community
1
OSUM-EChat
File size: 4,726 Bytes 
841f290
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
 
# Copyright (c) 2020 Mobvoi Inc. (authors: Binbin Zhang)
#
# 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 logging
import os
import re

import yaml
import torch
from collections import OrderedDict

import datetime


def filter_state_dict(model_state_dict, checkpoint_state_dict):
    filtered_state_dict = {}
    for key in checkpoint_state_dict:
        if key in model_state_dict:
            if model_state_dict[key].shape == checkpoint_state_dict[key].shape:
                filtered_state_dict[key] = checkpoint_state_dict[key]
            else:
                print(f"Skipping key '{key}' due to shape mismatch.")
    return filtered_state_dict


def load_checkpoint(model: torch.nn.Module, path: str) -> dict:
    rank = int(os.environ.get('RANK', 0))
    logging.info('[Rank {}] Checkpoint: loading from checkpoint {}'.format(
        rank, path))
    checkpoint = torch.load(path, map_location='cpu')
    filtered_checkpoint = filter_state_dict(model.state_dict(), checkpoint)
    missing_keys, unexpected_keys = model.load_state_dict(filtered_checkpoint,
                                                          strict=False)
    for key in missing_keys:
        logging.info("missing tensor: {}".format(key))
    for key in unexpected_keys:
        logging.info("unexpected tensor: {}".format(key))
    info_path = re.sub('.pt$', '.yaml', path)
    configs = {}
    if os.path.exists(info_path):
        with open(info_path, 'r') as fin:
            configs = yaml.load(fin, Loader=yaml.FullLoader)
        if configs is None:
            configs = {}
    return configs


def save_state_dict_and_infos(state_dict, path: str, infos=None):
    rank = int(os.environ.get('RANK', 0))
    logging.info('[Rank {}] Checkpoint: save to checkpoint {}'.format(
        rank, path))
    torch.save(state_dict, path)
    info_path = re.sub('.pt$', '.yaml', path)
    if infos is None:
        infos = {}
    infos['save_time'] = datetime.datetime.now().strftime('%d/%m/%Y %H:%M:%S')
    with open(info_path, 'w') as fout:
        data = yaml.dump(infos)
        fout.write(data)


def save_checkpoint(model: torch.nn.Module, path: str, infos=None):
    '''
    Args:
        infos (dict or None): any info you want to save.
    '''
    if isinstance(model, torch.nn.DataParallel):
        state_dict = model.module.state_dict()
    elif isinstance(model, torch.nn.parallel.DistributedDataParallel):
        state_dict = model.module.state_dict()
    else:
        state_dict = model.state_dict()
    save_state_dict_and_infos(state_dict, path, infos)


def filter_modules(model_state_dict, modules):
    rank = int(os.environ.get('RANK', 0))
    new_mods = []
    incorrect_mods = []
    mods_model = model_state_dict.keys()
    for mod in modules:
        if any(key.startswith(mod) for key in mods_model):
            new_mods += [mod]
        else:
            incorrect_mods += [mod]
    if incorrect_mods and rank == 0:
        logging.warning(
            "module(s) %s don't match or (partially match) "
            "available modules in model.",
            incorrect_mods,
        )
        logging.warning("for information, the existing modules in model are:")
        logging.warning("%s", mods_model)

    return new_mods


def load_trained_modules(model: torch.nn.Module, args: None):
    # Load encoder modules with pre-trained model(s).
    enc_model_path = args.enc_init
    enc_modules = args.enc_init_mods
    main_state_dict = model.state_dict()
    logging.warning("model(s) found for pre-initialization")
    if os.path.isfile(enc_model_path):
        logging.info('Checkpoint: loading from checkpoint %s for CPU' %
                     enc_model_path)
        model_state_dict = torch.load(enc_model_path, map_location='cpu')
        modules = filter_modules(model_state_dict, enc_modules)
        partial_state_dict = OrderedDict()
        for key, value in model_state_dict.items():
            if any(key.startswith(m) for m in modules):
                partial_state_dict[key] = value
        main_state_dict.update(partial_state_dict)
    else:
        logging.warning("model was not found : %s", enc_model_path)

    model.load_state_dict(main_state_dict)
    configs = {}
    return configs