ONNX
File size: 5,123 Bytes
0ed065b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ad0892
 
 
0ed065b
 
 
 
 
 
 
 
 
8ad0892
 
 
 
 
0ed065b
 
 
 
 
 
8ad0892
0ed065b
 
 
 
 
 
 
 
 
 
 
 
 
8ad0892
 
 
 
 
0ed065b
 
 
 
 
 
8ad0892
0ed065b
 
 
 
8ad0892
0ed065b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ad0892
 
 
 
 
 
 
 
 
 
 
0ed065b
 
8ad0892
0ed065b
8ad0892
0ed065b
 
8ad0892
0ed065b
 
 
 
 
 
 
 
 
 
 
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
141
142
143
144
145
146
147
148
149
150
151
152
153
import uform
import torch
import coremltools as ct
from os.path import join
from argparse import ArgumentParser

class TextEncoder(torch.nn.Module):
    def __init__(self, model):
        super().__init__()
        self.model = model.eval()
        
    def forward(self, input_ids, attention_mask):
        features = self.model.forward_features(
            input_ids, attention_mask
        )

        embeddings = self.model.forward_embedding(
            features, attention_mask
        )

        return features, embeddings
    
    
class ImageEncoder(torch.nn.Module):
    def __init__(self, model):
        super().__init__()
        self.model = model.eval()
        
    def forward(self, image):
        features = self.model.forward_features(
            image
        )

        embeddings = self.model.forward_embedding(
            features
        )

        return features, embeddings

def convert_model(opts):
    src_model = uform.get_model(opts.model_name)
    input_ids = torch.ones(1, src_model.text_encoder.max_position_embeddings, dtype=torch.int32)
    attention_mask = torch.ones(1, src_model.text_encoder.max_position_embeddings, dtype=torch.int32)
    image = torch.ones(1, 3, src_model.image_encoder.image_size, src_model.image_encoder.image_size, dtype=torch.float32)

    print('Tracing models…')
    image_encoder = ImageEncoder(src_model.image_encoder).eval()
    image_encoder = torch.jit.trace(image_encoder, image)
    text_encoder = TextEncoder(src_model.text_encoder).eval()
    text_encoder = torch.jit.trace(text_encoder, (input_ids, attention_mask))

    print('Converting models…')

    if opts.image_batchsize_lb == opts.image_batchsize_ub:
        image_batch_dim_shape = opts.image_batchsize_lb 
    else:
        image_batch_dim_shape = ct.RangeDim(lower_bound=opts.image_batchsize_lb, upper_bound=opts.image_batchsize_ub, default=1)

    image_encoder = ct.convert(
        image_encoder,
        convert_to='mlprogram',
        inputs=[
            ct.TensorType(
                name='image',
                shape=(image_batch_dim_shape,) + image.shape[1:],
                dtype=image.numpy().dtype
            )],
        outputs=[
            ct.TensorType(
                name='features'
            ),
            ct.TensorType(
                name='embeddings'
            )
        ],
        compute_precision=ct.precision.FLOAT16 if opts.use_fp16 else ct.precision.FLOAT32
    )

    if opts.text_batchsize_lb == opts.text_batchsize_ub:
        text_batch_dim_shape = opts.text_batchsize_lb 
    else:
        text_batch_dim_shape = ct.RangeDim(lower_bound=opts.text_batchsize_lb, upper_bound=opts.text_batchsize_ub, default=1)

    text_encoder = ct.convert(
        text_encoder,
        convert_to='mlprogram',
        inputs=[
            ct.TensorType(
                name='input_ids',
                shape=(text_batch_dim_shape,) + input_ids.shape[1:],
                dtype=input_ids.numpy().dtype
            ),
            ct.TensorType(
                name='attention_mask',
                shape=(text_batch_dim_shape,) + attention_mask.shape[1:],
                dtype=attention_mask.numpy().dtype
            )],
        outputs=[
            ct.TensorType(
                name="features"
            ),
            ct.TensorType(
                name="embeddings"
            )
        ],
        compute_precision=ct.precision.FLOAT16 if opts.use_fp16 else ct.precision.FLOAT32
    )

    print('Image encoder:', image_encoder, sep='\n')
    print('Text encoder:', text_encoder, sep='\n')

    image_encoder.save(join(opts.output_dir, f"{opts.model_name.replace('/', '.')}.image-encoder.mlpackage"))
    text_encoder.save(join(opts.output_dir, f"{opts.model_name.replace('/', '.')}.text-encoder.mlpackage"))


if __name__ == '__main__':
    opts = ArgumentParser()
    opts.add_argument('--model_name',
                      action='store',
                      type=str,
                      help='UForm model name')

    opts.add_argument('--text_batchsize_lb',
                      action='store',
                      type=int,
                      help='lower bound of batch size for text encoder')

    opts.add_argument('--text_batchsize_ub',
                      action='store',
                      type=int,
                      help='upper bound of batch size for text encoder')
    
    opts.add_argument('--image_batchsize_lb',
                      action='store',
                      type=int,
                      help='lower bound of batch size for image encoder')

    opts.add_argument('--image_batchsize_ub',
                      action='store',
                      type=int,
                      help='upper bound of batch size for image encoder')
    
    opts.add_argument('-use_fp16',
                      action='store_true',
                      help='whether to use fp16 for inference or not')
    
    opts.add_argument('--output_dir',
                      action='store',
                      type=str,
                      help='ouput directory')

    convert_model(opts.parse_args())