Upload folder using huggingface_hub

.ipynb_checkpoints/config-checkpoint.py

@@ -0,0 +1,42 @@
+from transformers import PretrainedConfig
+
+class CSDConfig(PretrainedConfig):
+    model_type = "CSDModel"
+    
+    def __init__(
+        self,
+        attention_dropout:float=0.0,
+        dropout:float=0.0,
+        hidden_act:str= "quick_gelu",
+        hidden_size:int= 1024,
+        image_size:int= 224,
+        initializer_factor:float= 1.0,
+        initializer_range:float=0.02,
+        intermediate_size:int=4096,
+        layer_norm_eps:float=1e-05,
+        num_attention_heads:int=16,
+        num_channels:int=3,
+        num_hidden_layers:int=24,
+        patch_size:int= 14,
+        projection_dim:int=768,
+        style_projection_dim:int=768,
+        content_projection_dim:int=768,
+        **kwargs,
+    ):   
+        super().__init__(**kwargs)
+        self.attention_dropout=attention_dropout
+        self.dropout=dropout
+        self.hidden_act=hidden_act
+        self.hidden_size=hidden_size
+        self.image_size=image_size
+        self.initializer_factor=initializer_factor
+        self.initializer_range=initializer_range
+        self.intermediate_size=intermediate_size
+        self.layer_norm_eps=layer_norm_eps
+        self.num_attention_heads=num_attention_heads
+        self.num_channels=num_channels
+        self.num_hidden_layers=num_hidden_layers
+        self.patch_size=patch_size
+        self.projection_dim=projection_dim
+        self.style_projection_dim=style_projection_dim
+        self.content_projection_dim=content_projection_dim

.ipynb_checkpoints/model-checkpoint.py

@@ -0,0 +1,16 @@
+import torch.nn as nn
+from .config import CSDConfig
+from transformers import PreTrainedModel, CLIPVisionModel
+
+class CSDModel(PreTrainedModel):
+    config_class = CSDConfig
+    def __init__(self, config: CSDConfig):
+        super().__init__(config)
+        self.backbone = CLIPVisionModel(config)
+        self.out_style = nn.Linear(config.hidden_size, config.style_projection_dim, bias=False)
+        self.out_content = nn.Linear(config.hidden_size, config.content_projection_dim, bias=False)
+
+    def forward(self, pixel_values):
+        features = self.backbone(pixel_values)
+        style_embeds = self.out_style(features)
+        content_embeds = self.out_content(features)

config.json

@@ -1,8 +1,13 @@
 {
+  "_name_or_path": "./",
   "architectures": [
     "CSDModel"
   ],
   "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "config.CSDConfig",
+    "AutoModel": "model.CSDModel"
+  },
   "content_projection_dim": 768,
   "dropout": 0.0,
   "hidden_act": "quick_gelu",
@@ -12,6 +17,7 @@
   "initializer_range": 0.02,
   "intermediate_size": 4096,
   "layer_norm_eps": 1e-05,
+  "model_type": "CSDModel",
   "num_attention_heads": 16,
   "num_channels": 3,
   "num_hidden_layers": 24,

config.py

@@ -0,0 +1,42 @@
+from transformers import PretrainedConfig
+
+class CSDConfig(PretrainedConfig):
+    model_type = "CSDModel"
+    
+    def __init__(
+        self,
+        attention_dropout:float=0.0,
+        dropout:float=0.0,
+        hidden_act:str= "quick_gelu",
+        hidden_size:int= 1024,
+        image_size:int= 224,
+        initializer_factor:float= 1.0,
+        initializer_range:float=0.02,
+        intermediate_size:int=4096,
+        layer_norm_eps:float=1e-05,
+        num_attention_heads:int=16,
+        num_channels:int=3,
+        num_hidden_layers:int=24,
+        patch_size:int= 14,
+        projection_dim:int=768,
+        style_projection_dim:int=768,
+        content_projection_dim:int=768,
+        **kwargs,
+    ):   
+        super().__init__(**kwargs)
+        self.attention_dropout=attention_dropout
+        self.dropout=dropout
+        self.hidden_act=hidden_act
+        self.hidden_size=hidden_size
+        self.image_size=image_size
+        self.initializer_factor=initializer_factor
+        self.initializer_range=initializer_range
+        self.intermediate_size=intermediate_size
+        self.layer_norm_eps=layer_norm_eps
+        self.num_attention_heads=num_attention_heads
+        self.num_channels=num_channels
+        self.num_hidden_layers=num_hidden_layers
+        self.patch_size=patch_size
+        self.projection_dim=projection_dim
+        self.style_projection_dim=style_projection_dim
+        self.content_projection_dim=content_projection_dim

model.py

@@ -0,0 +1,16 @@
+import torch.nn as nn
+from .config import CSDConfig
+from transformers import PreTrainedModel, CLIPVisionModel
+
+class CSDModel(PreTrainedModel):
+    config_class = CSDConfig
+    def __init__(self, config: CSDConfig):
+        super().__init__(config)
+        self.backbone = CLIPVisionModel(config)
+        self.out_style = nn.Linear(config.hidden_size, config.style_projection_dim, bias=False)
+        self.out_content = nn.Linear(config.hidden_size, config.content_projection_dim, bias=False)
+
+    def forward(self, pixel_values):
+        features = self.backbone(pixel_values)
+        style_embeds = self.out_style(features)
+        content_embeds = self.out_content(features)