Upload folder using huggingface_hub
Browse files- config.json +26 -0
- configuration_aimv2.py +62 -0
- model.safetensors +3 -0
- model.safetensors.index.json +180 -0
- modeling_aimv2.py +191 -0
- preprocessor_config.json +27 -0
config.json
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "checkpoints/aimv2-3B-patch14-448",
|
3 |
+
"architectures": [
|
4 |
+
"AIMv2Model"
|
5 |
+
],
|
6 |
+
"attention_dropout": 0.0,
|
7 |
+
"auto_map": {
|
8 |
+
"AutoConfig": "configuration_aimv2.AIMv2Config",
|
9 |
+
"AutoModel": "modeling_aimv2.AIMv2Model",
|
10 |
+
"FlaxAutoModel": "modeling_flax_aimv2.FlaxAIMv2Model"
|
11 |
+
},
|
12 |
+
"hidden_size": 3072,
|
13 |
+
"image_size": 448,
|
14 |
+
"intermediate_size": 8192,
|
15 |
+
"model_type": "aimv2",
|
16 |
+
"num_attention_heads": 24,
|
17 |
+
"num_channels": 3,
|
18 |
+
"num_hidden_layers": 24,
|
19 |
+
"patch_size": 14,
|
20 |
+
"projection_dropout": 0.0,
|
21 |
+
"qkv_bias": false,
|
22 |
+
"rms_norm_eps": 1e-05,
|
23 |
+
"torch_dtype": "bfloat16",
|
24 |
+
"transformers_version": "4.46.2",
|
25 |
+
"use_bias": false
|
26 |
+
}
|
configuration_aimv2.py
ADDED
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Any
|
2 |
+
|
3 |
+
from transformers.configuration_utils import PretrainedConfig
|
4 |
+
|
5 |
+
__all__ = ["AIMv2Config"]
|
6 |
+
|
7 |
+
|
8 |
+
class AIMv2Config(PretrainedConfig):
|
9 |
+
"""This is the configuration class to store the configuration of an [`AIMv2Model`].
|
10 |
+
|
11 |
+
Instantiating a configuration with the defaults will yield a similar configuration
|
12 |
+
to that of the [apple/aimv2-large-patch14-224](https://huggingface.co/apple/aimv2-large-patch14-224).
|
13 |
+
|
14 |
+
Args:
|
15 |
+
hidden_size: Dimension of the hidden representations.
|
16 |
+
intermediate_size: Dimension of the SwiGLU representations.
|
17 |
+
num_hidden_layers: Number of hidden layers in the Transformer.
|
18 |
+
num_attention_heads: Number of attention heads for each attention layer
|
19 |
+
in the Transformer.
|
20 |
+
num_channels: Number of input channels.
|
21 |
+
image_size: Image size.
|
22 |
+
patch_size: Patch size.
|
23 |
+
rms_norm_eps: Epsilon value used for the RMS normalization layer.
|
24 |
+
attention_dropout: Dropout ratio for attention probabilities.
|
25 |
+
projection_dropout: Dropout ratio for the projection layer after the attention.
|
26 |
+
qkv_bias: Whether to add a bias to the queries, keys and values.
|
27 |
+
use_bias: Whether to add a bias in the feed-forward and projection layers.
|
28 |
+
kwargs: Keyword arguments for the [`PretrainedConfig`].
|
29 |
+
"""
|
30 |
+
|
31 |
+
model_type: str = "aimv2"
|
32 |
+
|
33 |
+
def __init__(
|
34 |
+
self,
|
35 |
+
hidden_size: int = 1024,
|
36 |
+
intermediate_size: int = 2816,
|
37 |
+
num_hidden_layers: int = 24,
|
38 |
+
num_attention_heads: int = 8,
|
39 |
+
num_channels: int = 3,
|
40 |
+
image_size: int = 224,
|
41 |
+
patch_size: int = 14,
|
42 |
+
rms_norm_eps: float = 1e-5,
|
43 |
+
attention_dropout: float = 0.0,
|
44 |
+
projection_dropout: float = 0.0,
|
45 |
+
qkv_bias: bool = False,
|
46 |
+
use_bias: bool = False,
|
47 |
+
**kwargs: Any,
|
48 |
+
):
|
49 |
+
super().__init__(**kwargs)
|
50 |
+
self.hidden_size = hidden_size
|
51 |
+
self.intermediate_size = intermediate_size
|
52 |
+
self.num_hidden_layers = num_hidden_layers
|
53 |
+
self.num_attention_heads = num_attention_heads
|
54 |
+
self.num_channels = num_channels
|
55 |
+
self.patch_size = patch_size
|
56 |
+
self.image_size = image_size
|
57 |
+
self.attention_dropout = attention_dropout
|
58 |
+
self.rms_norm_eps = rms_norm_eps
|
59 |
+
|
60 |
+
self.projection_dropout = projection_dropout
|
61 |
+
self.qkv_bias = qkv_bias
|
62 |
+
self.use_bias = use_bias
|
model.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f445eaeab8c48ae50ab0de0157b47747b484064ed73e770e64db09eabc93927a
|
3 |
+
size 5446053960
|
model.safetensors.index.json
ADDED
@@ -0,0 +1,180 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"metadata": {
|
3 |
+
"total_size": 5446035456
|
4 |
+
},
|
5 |
+
"weight_map": {
|
6 |
+
"preprocessor.patchifier.norm.weight": "model-00001-of-00002.safetensors",
|
7 |
+
"preprocessor.patchifier.proj.bias": "model-00001-of-00002.safetensors",
|
8 |
+
"preprocessor.patchifier.proj.weight": "model-00001-of-00002.safetensors",
|
9 |
+
"preprocessor.pos_embed": "model-00001-of-00002.safetensors",
|
10 |
+
"trunk.blocks.0.attn.proj.weight": "model-00001-of-00002.safetensors",
|
11 |
+
"trunk.blocks.0.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
12 |
+
"trunk.blocks.0.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
13 |
+
"trunk.blocks.0.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
14 |
+
"trunk.blocks.0.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
15 |
+
"trunk.blocks.0.norm_1.weight": "model-00001-of-00002.safetensors",
|
16 |
+
"trunk.blocks.0.norm_2.weight": "model-00001-of-00002.safetensors",
|
17 |
+
"trunk.blocks.1.attn.proj.weight": "model-00001-of-00002.safetensors",
|
18 |
+
"trunk.blocks.1.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
19 |
+
"trunk.blocks.1.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
20 |
+
"trunk.blocks.1.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
21 |
+
"trunk.blocks.1.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
22 |
+
"trunk.blocks.1.norm_1.weight": "model-00001-of-00002.safetensors",
|
23 |
+
"trunk.blocks.1.norm_2.weight": "model-00001-of-00002.safetensors",
|
24 |
+
"trunk.blocks.10.attn.proj.weight": "model-00001-of-00002.safetensors",
|
25 |
+
"trunk.blocks.10.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
26 |
+
"trunk.blocks.10.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
27 |
+
"trunk.blocks.10.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
28 |
+
"trunk.blocks.10.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
29 |
+
"trunk.blocks.10.norm_1.weight": "model-00001-of-00002.safetensors",
|
30 |
+
"trunk.blocks.10.norm_2.weight": "model-00001-of-00002.safetensors",
|
31 |
+
"trunk.blocks.11.attn.proj.weight": "model-00001-of-00002.safetensors",
|
32 |
+
"trunk.blocks.11.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
33 |
+
"trunk.blocks.11.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
34 |
+
"trunk.blocks.11.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
35 |
+
"trunk.blocks.11.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
36 |
+
"trunk.blocks.11.norm_1.weight": "model-00001-of-00002.safetensors",
|
37 |
+
"trunk.blocks.11.norm_2.weight": "model-00001-of-00002.safetensors",
|
38 |
+
"trunk.blocks.12.attn.proj.weight": "model-00001-of-00002.safetensors",
|
39 |
+
"trunk.blocks.12.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
40 |
+
"trunk.blocks.12.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
41 |
+
"trunk.blocks.12.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
42 |
+
"trunk.blocks.12.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
43 |
+
"trunk.blocks.12.norm_1.weight": "model-00001-of-00002.safetensors",
|
44 |
+
"trunk.blocks.12.norm_2.weight": "model-00001-of-00002.safetensors",
|
45 |
+
"trunk.blocks.13.attn.proj.weight": "model-00001-of-00002.safetensors",
|
46 |
+
"trunk.blocks.13.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
47 |
+
"trunk.blocks.13.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
48 |
+
"trunk.blocks.13.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
49 |
+
"trunk.blocks.13.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
50 |
+
"trunk.blocks.13.norm_1.weight": "model-00001-of-00002.safetensors",
|
51 |
+
"trunk.blocks.13.norm_2.weight": "model-00001-of-00002.safetensors",
|
52 |
+
"trunk.blocks.14.attn.proj.weight": "model-00001-of-00002.safetensors",
|
53 |
+
"trunk.blocks.14.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
54 |
+
"trunk.blocks.14.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
55 |
+
"trunk.blocks.14.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
56 |
+
"trunk.blocks.14.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
57 |
+
"trunk.blocks.14.norm_1.weight": "model-00001-of-00002.safetensors",
|
58 |
+
"trunk.blocks.14.norm_2.weight": "model-00001-of-00002.safetensors",
|
59 |
+
"trunk.blocks.15.attn.proj.weight": "model-00001-of-00002.safetensors",
|
60 |
+
"trunk.blocks.15.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
61 |
+
"trunk.blocks.15.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
62 |
+
"trunk.blocks.15.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
63 |
+
"trunk.blocks.15.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
64 |
+
"trunk.blocks.15.norm_1.weight": "model-00001-of-00002.safetensors",
|
65 |
+
"trunk.blocks.15.norm_2.weight": "model-00001-of-00002.safetensors",
|
66 |
+
"trunk.blocks.16.attn.proj.weight": "model-00001-of-00002.safetensors",
|
67 |
+
"trunk.blocks.16.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
68 |
+
"trunk.blocks.16.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
69 |
+
"trunk.blocks.16.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
70 |
+
"trunk.blocks.16.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
71 |
+
"trunk.blocks.16.norm_1.weight": "model-00001-of-00002.safetensors",
|
72 |
+
"trunk.blocks.16.norm_2.weight": "model-00001-of-00002.safetensors",
|
73 |
+
"trunk.blocks.17.attn.proj.weight": "model-00001-of-00002.safetensors",
|
74 |
+
"trunk.blocks.17.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
75 |
+
"trunk.blocks.17.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
76 |
+
"trunk.blocks.17.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
77 |
+
"trunk.blocks.17.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
78 |
+
"trunk.blocks.17.norm_1.weight": "model-00001-of-00002.safetensors",
|
79 |
+
"trunk.blocks.17.norm_2.weight": "model-00001-of-00002.safetensors",
|
80 |
+
"trunk.blocks.18.attn.proj.weight": "model-00001-of-00002.safetensors",
|
81 |
+
"trunk.blocks.18.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
82 |
+
"trunk.blocks.18.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
83 |
+
"trunk.blocks.18.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
84 |
+
"trunk.blocks.18.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
85 |
+
"trunk.blocks.18.norm_1.weight": "model-00001-of-00002.safetensors",
|
86 |
+
"trunk.blocks.18.norm_2.weight": "model-00001-of-00002.safetensors",
|
87 |
+
"trunk.blocks.19.attn.proj.weight": "model-00001-of-00002.safetensors",
|
88 |
+
"trunk.blocks.19.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
89 |
+
"trunk.blocks.19.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
90 |
+
"trunk.blocks.19.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
91 |
+
"trunk.blocks.19.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
92 |
+
"trunk.blocks.19.norm_1.weight": "model-00001-of-00002.safetensors",
|
93 |
+
"trunk.blocks.19.norm_2.weight": "model-00001-of-00002.safetensors",
|
94 |
+
"trunk.blocks.2.attn.proj.weight": "model-00001-of-00002.safetensors",
|
95 |
+
"trunk.blocks.2.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
96 |
+
"trunk.blocks.2.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
97 |
+
"trunk.blocks.2.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
98 |
+
"trunk.blocks.2.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
99 |
+
"trunk.blocks.2.norm_1.weight": "model-00001-of-00002.safetensors",
|
100 |
+
"trunk.blocks.2.norm_2.weight": "model-00001-of-00002.safetensors",
|
101 |
+
"trunk.blocks.20.attn.proj.weight": "model-00001-of-00002.safetensors",
|
102 |
+
"trunk.blocks.20.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
103 |
+
"trunk.blocks.20.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
104 |
+
"trunk.blocks.20.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
105 |
+
"trunk.blocks.20.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
106 |
+
"trunk.blocks.20.norm_1.weight": "model-00001-of-00002.safetensors",
|
107 |
+
"trunk.blocks.20.norm_2.weight": "model-00001-of-00002.safetensors",
|
108 |
+
"trunk.blocks.21.attn.proj.weight": "model-00001-of-00002.safetensors",
|
109 |
+
"trunk.blocks.21.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
110 |
+
"trunk.blocks.21.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
111 |
+
"trunk.blocks.21.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
112 |
+
"trunk.blocks.21.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
113 |
+
"trunk.blocks.21.norm_1.weight": "model-00001-of-00002.safetensors",
|
114 |
+
"trunk.blocks.21.norm_2.weight": "model-00001-of-00002.safetensors",
|
115 |
+
"trunk.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
|
116 |
+
"trunk.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
117 |
+
"trunk.blocks.22.mlp.fc1.weight": "model-00002-of-00002.safetensors",
|
118 |
+
"trunk.blocks.22.mlp.fc2.weight": "model-00002-of-00002.safetensors",
|
119 |
+
"trunk.blocks.22.mlp.fc3.weight": "model-00002-of-00002.safetensors",
|
120 |
+
"trunk.blocks.22.norm_1.weight": "model-00002-of-00002.safetensors",
|
121 |
+
"trunk.blocks.22.norm_2.weight": "model-00002-of-00002.safetensors",
|
122 |
+
"trunk.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
|
123 |
+
"trunk.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
124 |
+
"trunk.blocks.23.mlp.fc1.weight": "model-00002-of-00002.safetensors",
|
125 |
+
"trunk.blocks.23.mlp.fc2.weight": "model-00002-of-00002.safetensors",
|
126 |
+
"trunk.blocks.23.mlp.fc3.weight": "model-00002-of-00002.safetensors",
|
127 |
+
"trunk.blocks.23.norm_1.weight": "model-00002-of-00002.safetensors",
|
128 |
+
"trunk.blocks.23.norm_2.weight": "model-00002-of-00002.safetensors",
|
129 |
+
"trunk.blocks.3.attn.proj.weight": "model-00001-of-00002.safetensors",
|
130 |
+
"trunk.blocks.3.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
131 |
+
"trunk.blocks.3.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
132 |
+
"trunk.blocks.3.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
133 |
+
"trunk.blocks.3.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
134 |
+
"trunk.blocks.3.norm_1.weight": "model-00001-of-00002.safetensors",
|
135 |
+
"trunk.blocks.3.norm_2.weight": "model-00001-of-00002.safetensors",
|
136 |
+
"trunk.blocks.4.attn.proj.weight": "model-00001-of-00002.safetensors",
|
137 |
+
"trunk.blocks.4.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
138 |
+
"trunk.blocks.4.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
139 |
+
"trunk.blocks.4.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
140 |
+
"trunk.blocks.4.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
141 |
+
"trunk.blocks.4.norm_1.weight": "model-00001-of-00002.safetensors",
|
142 |
+
"trunk.blocks.4.norm_2.weight": "model-00001-of-00002.safetensors",
|
143 |
+
"trunk.blocks.5.attn.proj.weight": "model-00001-of-00002.safetensors",
|
144 |
+
"trunk.blocks.5.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
145 |
+
"trunk.blocks.5.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
146 |
+
"trunk.blocks.5.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
147 |
+
"trunk.blocks.5.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
148 |
+
"trunk.blocks.5.norm_1.weight": "model-00001-of-00002.safetensors",
|
149 |
+
"trunk.blocks.5.norm_2.weight": "model-00001-of-00002.safetensors",
|
150 |
+
"trunk.blocks.6.attn.proj.weight": "model-00001-of-00002.safetensors",
|
151 |
+
"trunk.blocks.6.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
152 |
+
"trunk.blocks.6.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
153 |
+
"trunk.blocks.6.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
154 |
+
"trunk.blocks.6.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
155 |
+
"trunk.blocks.6.norm_1.weight": "model-00001-of-00002.safetensors",
|
156 |
+
"trunk.blocks.6.norm_2.weight": "model-00001-of-00002.safetensors",
|
157 |
+
"trunk.blocks.7.attn.proj.weight": "model-00001-of-00002.safetensors",
|
158 |
+
"trunk.blocks.7.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
159 |
+
"trunk.blocks.7.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
160 |
+
"trunk.blocks.7.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
161 |
+
"trunk.blocks.7.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
162 |
+
"trunk.blocks.7.norm_1.weight": "model-00001-of-00002.safetensors",
|
163 |
+
"trunk.blocks.7.norm_2.weight": "model-00001-of-00002.safetensors",
|
164 |
+
"trunk.blocks.8.attn.proj.weight": "model-00001-of-00002.safetensors",
|
165 |
+
"trunk.blocks.8.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
166 |
+
"trunk.blocks.8.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
167 |
+
"trunk.blocks.8.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
168 |
+
"trunk.blocks.8.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
169 |
+
"trunk.blocks.8.norm_1.weight": "model-00001-of-00002.safetensors",
|
170 |
+
"trunk.blocks.8.norm_2.weight": "model-00001-of-00002.safetensors",
|
171 |
+
"trunk.blocks.9.attn.proj.weight": "model-00001-of-00002.safetensors",
|
172 |
+
"trunk.blocks.9.attn.qkv.weight": "model-00001-of-00002.safetensors",
|
173 |
+
"trunk.blocks.9.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
174 |
+
"trunk.blocks.9.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
175 |
+
"trunk.blocks.9.mlp.fc3.weight": "model-00001-of-00002.safetensors",
|
176 |
+
"trunk.blocks.9.norm_1.weight": "model-00001-of-00002.safetensors",
|
177 |
+
"trunk.blocks.9.norm_2.weight": "model-00001-of-00002.safetensors",
|
178 |
+
"trunk.post_trunk_norm.weight": "model-00002-of-00002.safetensors"
|
179 |
+
}
|
180 |
+
}
|
modeling_aimv2.py
ADDED
@@ -0,0 +1,191 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Optional, Tuple, Union
|
2 |
+
|
3 |
+
import torch
|
4 |
+
from .configuration_aimv2 import AIMv2Config
|
5 |
+
from torch import nn
|
6 |
+
from torch.nn import functional as F
|
7 |
+
from transformers.modeling_outputs import BaseModelOutputWithNoAttention
|
8 |
+
from transformers.modeling_utils import PreTrainedModel
|
9 |
+
|
10 |
+
__all__ = ["AIMv2Model"]
|
11 |
+
|
12 |
+
|
13 |
+
class RMSNorm(nn.Module):
|
14 |
+
def __init__(self, dim: int, eps: float = 1e-6):
|
15 |
+
super().__init__()
|
16 |
+
self.weight = nn.Parameter(torch.ones(dim))
|
17 |
+
self.eps = eps
|
18 |
+
|
19 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
20 |
+
output = self._norm(x.float()).type_as(x)
|
21 |
+
return output * self.weight
|
22 |
+
|
23 |
+
def extra_repr(self) -> str:
|
24 |
+
return f"{tuple(self.weight.shape)}, eps={self.eps}"
|
25 |
+
|
26 |
+
def _norm(self, x: torch.Tensor) -> torch.Tensor:
|
27 |
+
return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
|
28 |
+
|
29 |
+
|
30 |
+
class AIMv2SwiGLUFFN(nn.Module):
|
31 |
+
def __init__(self, config: AIMv2Config):
|
32 |
+
super().__init__()
|
33 |
+
hidden_features = config.intermediate_size
|
34 |
+
in_features = config.hidden_size
|
35 |
+
bias = config.use_bias
|
36 |
+
|
37 |
+
self.fc1 = nn.Linear(in_features, hidden_features, bias=bias)
|
38 |
+
self.fc2 = nn.Linear(hidden_features, in_features, bias=bias)
|
39 |
+
self.fc3 = nn.Linear(in_features, hidden_features, bias=bias)
|
40 |
+
|
41 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
42 |
+
x = F.silu(self.fc1(x)) * self.fc3(x)
|
43 |
+
x = self.fc2(x)
|
44 |
+
return x
|
45 |
+
|
46 |
+
|
47 |
+
class AIMv2PatchEmbed(nn.Module):
|
48 |
+
def __init__(self, config: AIMv2Config):
|
49 |
+
super().__init__()
|
50 |
+
self.proj = nn.Conv2d(
|
51 |
+
config.num_channels,
|
52 |
+
config.hidden_size,
|
53 |
+
kernel_size=(config.patch_size, config.patch_size),
|
54 |
+
stride=(config.patch_size, config.patch_size),
|
55 |
+
)
|
56 |
+
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
57 |
+
|
58 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
59 |
+
x = self.proj(x).flatten(2).transpose(1, 2)
|
60 |
+
x = self.norm(x)
|
61 |
+
return x
|
62 |
+
|
63 |
+
|
64 |
+
class AIMv2ViTPreprocessor(nn.Module):
|
65 |
+
def __init__(self, config: AIMv2Config):
|
66 |
+
super().__init__()
|
67 |
+
num_patches = (config.image_size // config.patch_size) ** 2
|
68 |
+
|
69 |
+
self.patchifier = AIMv2PatchEmbed(config)
|
70 |
+
self.pos_embed = nn.Parameter(torch.zeros((1, num_patches, config.hidden_size)))
|
71 |
+
|
72 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
73 |
+
tokens = self.patchifier(x)
|
74 |
+
_, N, _ = tokens.shape
|
75 |
+
pos_embed = self.pos_embed.to(tokens.device)
|
76 |
+
tokens = tokens + pos_embed[:, :N]
|
77 |
+
return tokens
|
78 |
+
|
79 |
+
|
80 |
+
class AIMv2Attention(nn.Module):
|
81 |
+
def __init__(self, config: AIMv2Config):
|
82 |
+
super().__init__()
|
83 |
+
dim = config.hidden_size
|
84 |
+
|
85 |
+
self.num_heads = config.num_attention_heads
|
86 |
+
self.qkv = nn.Linear(dim, dim * 3, bias=config.qkv_bias)
|
87 |
+
self.attn_drop = nn.Dropout(config.attention_dropout)
|
88 |
+
self.proj = nn.Linear(dim, dim, bias=config.use_bias)
|
89 |
+
self.proj_drop = nn.Dropout(config.projection_dropout)
|
90 |
+
|
91 |
+
def forward(
|
92 |
+
self, x: torch.Tensor, mask: Optional[torch.Tensor] = None
|
93 |
+
) -> torch.Tensor:
|
94 |
+
B, N, C = x.shape
|
95 |
+
qkv = (
|
96 |
+
self.qkv(x)
|
97 |
+
.reshape(B, N, 3, self.num_heads, C // self.num_heads)
|
98 |
+
.permute(2, 0, 3, 1, 4)
|
99 |
+
)
|
100 |
+
q, k, v = qkv.unbind(0)
|
101 |
+
|
102 |
+
x = F.scaled_dot_product_attention(q, k, v, attn_mask=mask)
|
103 |
+
x = x.transpose(1, 2).contiguous().reshape(B, N, C)
|
104 |
+
x = self.proj(x)
|
105 |
+
x = self.proj_drop(x)
|
106 |
+
return x
|
107 |
+
|
108 |
+
|
109 |
+
class AIMv2Block(nn.Module):
|
110 |
+
def __init__(self, config: AIMv2Config):
|
111 |
+
super().__init__()
|
112 |
+
self.attn = AIMv2Attention(config)
|
113 |
+
self.norm_1 = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
114 |
+
self.mlp = AIMv2SwiGLUFFN(config)
|
115 |
+
self.norm_2 = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
116 |
+
|
117 |
+
def forward(
|
118 |
+
self, x: torch.Tensor, mask: Optional[torch.Tensor] = None
|
119 |
+
) -> torch.Tensor:
|
120 |
+
x = x + self.attn(self.norm_1(x), mask)
|
121 |
+
x = x + self.mlp(self.norm_2(x))
|
122 |
+
return x
|
123 |
+
|
124 |
+
|
125 |
+
class AIMv2Transformer(nn.Module):
|
126 |
+
def __init__(self, config: AIMv2Config):
|
127 |
+
super().__init__()
|
128 |
+
self.blocks = nn.ModuleList(
|
129 |
+
[AIMv2Block(config) for _ in range(config.num_hidden_layers)]
|
130 |
+
)
|
131 |
+
self.post_trunk_norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
132 |
+
|
133 |
+
def forward(
|
134 |
+
self,
|
135 |
+
tokens: torch.Tensor,
|
136 |
+
mask: Optional[torch.Tensor] = None,
|
137 |
+
output_hidden_states: bool = False,
|
138 |
+
) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor, ...]]]:
|
139 |
+
hidden_states = () if output_hidden_states else None
|
140 |
+
for block in self.blocks:
|
141 |
+
tokens = block(tokens, mask)
|
142 |
+
if output_hidden_states:
|
143 |
+
hidden_states += (tokens,)
|
144 |
+
tokens = self.post_trunk_norm(tokens)
|
145 |
+
return tokens, hidden_states
|
146 |
+
|
147 |
+
|
148 |
+
class AIMv2PretrainedModel(PreTrainedModel):
|
149 |
+
config_class = AIMv2Config
|
150 |
+
base_model_prefix = "aimv2"
|
151 |
+
main_input_name = "pixel_values"
|
152 |
+
_supports_sdpa = True
|
153 |
+
|
154 |
+
|
155 |
+
class AIMv2Model(AIMv2PretrainedModel):
|
156 |
+
def __init__(self, config: AIMv2Config):
|
157 |
+
super().__init__(config)
|
158 |
+
self.preprocessor = AIMv2ViTPreprocessor(config)
|
159 |
+
self.trunk = AIMv2Transformer(config)
|
160 |
+
|
161 |
+
def forward(
|
162 |
+
self,
|
163 |
+
pixel_values: torch.Tensor,
|
164 |
+
mask: Optional[torch.Tensor] = None,
|
165 |
+
output_hidden_states: Optional[bool] = None,
|
166 |
+
return_dict: Optional[bool] = None,
|
167 |
+
) -> Union[
|
168 |
+
Tuple[torch.Tensor],
|
169 |
+
Tuple[torch.Tensor, Tuple[torch.Tensor, ...]],
|
170 |
+
BaseModelOutputWithNoAttention,
|
171 |
+
]:
|
172 |
+
if output_hidden_states is None:
|
173 |
+
output_hidden_states = self.config.output_hidden_states
|
174 |
+
if return_dict is None:
|
175 |
+
return_dict = self.config.use_return_dict
|
176 |
+
|
177 |
+
x = self.preprocessor(pixel_values)
|
178 |
+
x, hidden_states = self.trunk(
|
179 |
+
x, mask, output_hidden_states=output_hidden_states
|
180 |
+
)
|
181 |
+
|
182 |
+
if not return_dict:
|
183 |
+
res = (x,)
|
184 |
+
res += (hidden_states,) if output_hidden_states else ()
|
185 |
+
return res
|
186 |
+
|
187 |
+
return BaseModelOutputWithNoAttention(
|
188 |
+
last_hidden_state=x,
|
189 |
+
hidden_states=hidden_states,
|
190 |
+
)
|
191 |
+
|
preprocessor_config.json
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"crop_size": {
|
3 |
+
"height": 448,
|
4 |
+
"width": 448
|
5 |
+
},
|
6 |
+
"do_center_crop": true,
|
7 |
+
"do_convert_rgb": true,
|
8 |
+
"do_normalize": true,
|
9 |
+
"do_rescale": true,
|
10 |
+
"do_resize": true,
|
11 |
+
"image_mean": [
|
12 |
+
0.48145466,
|
13 |
+
0.4578275,
|
14 |
+
0.40821073
|
15 |
+
],
|
16 |
+
"image_processor_type": "CLIPImageProcessor",
|
17 |
+
"image_std": [
|
18 |
+
0.26862954,
|
19 |
+
0.26130258,
|
20 |
+
0.27577711
|
21 |
+
],
|
22 |
+
"resample": 3,
|
23 |
+
"rescale_factor": 0.00392156862745098,
|
24 |
+
"size": {
|
25 |
+
"shortest_edge": 448
|
26 |
+
}
|
27 |
+
}
|