Upload model
Browse files- hf_model.py +19 -27
- model.py +54 -6
- pytorch_model.bin +2 -2
hf_model.py
CHANGED
@@ -20,6 +20,7 @@ from transformers import PretrainedConfig, PreTrainedModel
|
|
20 |
|
21 |
|
22 |
from .model import create_model_from_args
|
|
|
23 |
from .input_conditioner import get_default_conditioner, InputConditioner
|
24 |
|
25 |
|
@@ -42,7 +43,11 @@ class RADIOConfig(PretrainedConfig):
|
|
42 |
|
43 |
|
44 |
class RADIOModel(PreTrainedModel):
|
45 |
-
"""Pretrained Hugging Face model for RADIO.
|
|
|
|
|
|
|
|
|
46 |
|
47 |
config_class = RADIOConfig
|
48 |
|
@@ -52,32 +57,19 @@ class RADIOModel(PreTrainedModel):
|
|
52 |
RADIOArgs = namedtuple("RADIOArgs", config.args.keys())
|
53 |
args = RADIOArgs(**config.args)
|
54 |
self.config = config
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
def forward(self, x: torch.Tensor):
|
59 |
-
x = self.input_conditioner(x)
|
60 |
|
61 |
-
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
patch_gen = getattr(self.model, "patch_generator", None)
|
67 |
-
if patch_gen is not None:
|
68 |
-
summary = y[:, : patch_gen.num_cls_tokens].flatten(1)
|
69 |
-
all_feat = y[:, patch_gen.num_skip :]
|
70 |
-
elif self.model.global_pool == "avg":
|
71 |
-
summary = y[:, self.model.num_prefix_tokens :].mean(dim=1)
|
72 |
-
all_feat = y
|
73 |
-
else:
|
74 |
-
summary = y[:, 0]
|
75 |
-
all_feat = y[:, 1:]
|
76 |
-
else:
|
77 |
-
raise ValueError("Unsupported model type")
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
elif self.config.return_summary:
|
82 |
-
return summary
|
83 |
-
return all_feat
|
|
|
20 |
|
21 |
|
22 |
from .model import create_model_from_args
|
23 |
+
from .model import RADIOModel as RADIOModelBase
|
24 |
from .input_conditioner import get_default_conditioner, InputConditioner
|
25 |
|
26 |
|
|
|
43 |
|
44 |
|
45 |
class RADIOModel(PreTrainedModel):
|
46 |
+
"""Pretrained Hugging Face model for RADIO.
|
47 |
+
|
48 |
+
This classes inherits from both PreTrainedModel, which provides
|
49 |
+
HuggingFace's functionality for loading and saving models.
|
50 |
+
"""
|
51 |
|
52 |
config_class = RADIOConfig
|
53 |
|
|
|
57 |
RADIOArgs = namedtuple("RADIOArgs", config.args.keys())
|
58 |
args = RADIOArgs(**config.args)
|
59 |
self.config = config
|
60 |
+
model = create_model_from_args(args)
|
61 |
+
input_conditioner: InputConditioner = get_default_conditioner()
|
|
|
|
|
|
|
62 |
|
63 |
+
self.radio_model = RADIOModelBase(
|
64 |
+
model,
|
65 |
+
input_conditioner,
|
66 |
+
config.return_summary,
|
67 |
+
config.return_spatial_features,
|
68 |
+
)
|
69 |
|
70 |
+
@property
|
71 |
+
def model(self) -> VisionTransformer:
|
72 |
+
return self.radio_model.model
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
+
def forward(self, x: torch.Tensor):
|
75 |
+
return self.radio_model.forward(x)
|
|
|
|
|
|
model.py
CHANGED
@@ -6,11 +6,56 @@
|
|
6 |
# distribution of this software and related documentation without an express
|
7 |
# license agreement from NVIDIA CORPORATION is strictly prohibited.
|
8 |
|
|
|
9 |
from torch import nn
|
10 |
|
11 |
-
from timm.models import create_model
|
12 |
|
13 |
from .enable_cpe_support import enable_cpe
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
|
16 |
def create_model_from_args(args) -> nn.Module:
|
@@ -36,13 +81,16 @@ def create_model_from_args(args) -> nn.Module:
|
|
36 |
**args.model_kwargs,
|
37 |
)
|
38 |
|
39 |
-
assert
|
|
|
|
|
40 |
|
41 |
if args.cpe_max_size is not None:
|
42 |
-
enable_cpe(
|
43 |
-
|
44 |
-
|
45 |
-
|
|
|
46 |
)
|
47 |
|
48 |
return model
|
|
|
6 |
# distribution of this software and related documentation without an express
|
7 |
# license agreement from NVIDIA CORPORATION is strictly prohibited.
|
8 |
|
9 |
+
import torch
|
10 |
from torch import nn
|
11 |
|
12 |
+
from timm.models import create_model, VisionTransformer
|
13 |
|
14 |
from .enable_cpe_support import enable_cpe
|
15 |
+
from .input_conditioner import InputConditioner
|
16 |
+
|
17 |
+
|
18 |
+
class RADIOModel(nn.Module):
|
19 |
+
def __init__(
|
20 |
+
self,
|
21 |
+
model: nn.Module,
|
22 |
+
input_conditioner: InputConditioner,
|
23 |
+
return_summary: bool,
|
24 |
+
return_spatial_features: bool,
|
25 |
+
):
|
26 |
+
super().__init__()
|
27 |
+
|
28 |
+
self.model = model
|
29 |
+
self.input_conditioner = input_conditioner
|
30 |
+
self.return_summary = return_summary
|
31 |
+
self.return_spatial_features = return_spatial_features
|
32 |
+
|
33 |
+
def forward(self, x: torch.Tensor):
|
34 |
+
x = self.input_conditioner(x)
|
35 |
+
|
36 |
+
y = self.model.forward_features(x)
|
37 |
+
|
38 |
+
if isinstance(y, (list, tuple)):
|
39 |
+
summary, all_feat = y
|
40 |
+
elif isinstance(self.model, VisionTransformer):
|
41 |
+
patch_gen = getattr(self.model, "patch_generator", None)
|
42 |
+
if patch_gen is not None:
|
43 |
+
summary = y[:, : patch_gen.num_cls_tokens].flatten(1)
|
44 |
+
all_feat = y[:, patch_gen.num_skip :]
|
45 |
+
elif self.model.global_pool == "avg":
|
46 |
+
summary = y[:, self.model.num_prefix_tokens :].mean(dim=1)
|
47 |
+
all_feat = y
|
48 |
+
else:
|
49 |
+
summary = y[:, 0]
|
50 |
+
all_feat = y[:, 1:]
|
51 |
+
else:
|
52 |
+
raise ValueError("Unsupported model type")
|
53 |
+
|
54 |
+
if self.return_summary and self.return_spatial_features:
|
55 |
+
return summary, all_feat
|
56 |
+
elif self.return_summary:
|
57 |
+
return summary
|
58 |
+
return all_feat
|
59 |
|
60 |
|
61 |
def create_model_from_args(args) -> nn.Module:
|
|
|
81 |
**args.model_kwargs,
|
82 |
)
|
83 |
|
84 |
+
assert (
|
85 |
+
not args.cls_token_per_teacher or args.cpe_max_size is not None
|
86 |
+
), "CPE must be enabled for multiple CLS tokens!"
|
87 |
|
88 |
if args.cpe_max_size is not None:
|
89 |
+
enable_cpe(
|
90 |
+
model,
|
91 |
+
args.cpe_max_size,
|
92 |
+
num_cls_tokens=len(args.teachers) if args.cls_token_per_teacher else 1,
|
93 |
+
register_multiple=args.register_multiple,
|
94 |
)
|
95 |
|
96 |
return model
|
pytorch_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ad369b92359d9a42f93f6bbb9be2191f79b4b6fc923fdd31d992ca32336f608d
|
3 |
+
size 2662624177
|