recap/config_helpers.py at main · git-disl/recap · GitHub

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
132
133
134
135
136
137
138
139
140
from transformers.models.bert.modeling_bert import BertLayer
from transformers.models.vit.modeling_vit import ViTLayer
from transformers.models.swin.modeling_swin import SwinLayer


def get_prune_config_for_attn(args, model, prune_params_dict):
    sparse_ratio = prune_params_dict['sparse_ratio']
    max_sparse_ratio = prune_params_dict.get('max_sparse_ratio', 1)
    granularity = prune_params_dict['granularity']
    config_list = []

    if 'bert' in str(model.__class__):
        attention_qkv_str = '.attention.self*'
        attention_output_str = '.attention.output.dense'
        dep_id = -1
    elif 'vit' in str(model.__class__):
        attention_qkv_str = '.attention.attention*'
        attention_output_str = '.attention.output.dense'
        dep_id = -1
    elif 'mask2former' in str(model.__class__):
        attention_qkv_str = '.attention.self*'
        attention_output_str = '.attention.output.dense'
        dep_id = -3
    else:
        raise NotImplementedError

    for name, module in model.named_modules():

        if 'encoder' in name:
            inc = 0
        else:
            inc = 100

        if isinstance(module, SwinLayer):
            if 'm2f' in args.arch:
                if '.0.' in name:
                    granularity_ = [32, 128]
                elif '.1.' in name:
                    granularity_ = [32, 256]
                elif '.2.' in name:
                    granularity_ = [32, 512]
                else:
                    granularity_ = [32, 1024]
            else:
                if '.0.' in name:
                    granularity_ = [32, 192]
                elif '.1.' in name:
                    granularity_ = [32, 384]
                elif '.2.' in name:
                    granularity_ = [32, 768]
                else:
                    granularity_ = [32, 1536]
        else:
            granularity_ = granularity

        if isinstance(module, BertLayer) or isinstance(module, ViTLayer) or isinstance(module, SwinLayer):
            config_list.append({'op_types': ['Linear'],
                                'op_names_re': [f'{name}{attention_qkv_str}'],
                                'dependency_group_id': int(name.split('.')[dep_id]) + inc,
                                'sparse_ratio': sparse_ratio,
                                'max_sparse_ratio': max_sparse_ratio,
                                'granularity': granularity_,
                                'global_group_id': inc
                                })
            config_list.append({'op_names': [f'{name}{attention_output_str}'],
                                'dependency_group_id': int(name.split('.')[dep_id]) + inc,
                                'sparse_ratio': sparse_ratio,
                                'max_sparse_ratio': max_sparse_ratio,
                                'granularity': list(reversed(granularity_)),
                                'global_group_id': inc
                                })
    return config_list


def get_prune_config_for_ffn(args, model, prune_params_dict):
    sparse_ratio = prune_params_dict['sparse_ratio']
    max_sparse_ratio = prune_params_dict.get('max_sparse_ratio', 1)
    granularity = prune_params_dict['granularity']
    config_list = []

    if 'bert' in str(model.__class__):
        intermediate_str = '.intermediate.dense'
        output_str = '.output.dense'
        dep_id =  -1
    elif 'vit' in str(model.__class__):
        intermediate_str = '.intermediate.dense'
        output_str = '.output.dense'
        dep_id = -1
    elif 'mask2former' in str(model.__class__):
        intermediate_str = '.intermediate.dense'
        output_str = '.output.dense'
        dep_id = -3
    else:
        raise NotImplementedError

    for name, module in model.named_modules():

        if 'encoder' in name:
            inc = 200
        else:
            inc = 300

        if isinstance(module, SwinLayer):
            if 'm2f' in args.arch:
                if '.0.' in name:
                    granularity_ = [32, 128]
                elif '.1.' in name:
                    granularity_ = [32, 256]
                elif '.2.' in name:
                    granularity_ = [32, 512]
                else:
                    granularity_ = [32, 1024]
            else:
                if '.0.' in name:
                    granularity_ = [1, 192]
                elif '.1.' in name:
                    granularity_ = [1, 384]
                elif '.2.' in name:
                    granularity_ = [1, 768]
                else:
                    granularity_ = [1, 1536]
        else:
            granularity_ = granularity

        if isinstance(module, BertLayer) or isinstance(module, ViTLayer) or isinstance(module, SwinLayer):
            config_list.append({'op_names': [f'{name}{intermediate_str}'],
                                'dependency_group_id': int(name.split('.')[dep_id]) + inc,
                                'sparse_ratio': sparse_ratio,
                                'max_sparse_ratio': max_sparse_ratio,
                                'granularity': granularity_,
                                'global_group_id': inc
                                })
            config_list.append({'op_names': [f'{name}{output_str}'],
                                'dependency_group_id': int(name.split('.')[dep_id]) + inc,
                                'sparse_ratio': sparse_ratio,
                                'max_sparse_ratio': max_sparse_ratio,
                                'granularity': list(reversed(granularity_)),
                                'global_group_id': inc
                                })
    return config_list