Update to allow model to be loaded with transformers

.gitignore

@@ -0,0 +1 @@
+__pycache__

__init__.py

@@ -0,0 +1,10 @@
+from transformers import AutoConfig, AutoModel
+
+from .unet_config import UNetConfig
+from .unet_hf import UNetModel
+
+# Register with Hugging Face Auto classes
+AutoConfig.register("unet", UNetConfig)
+AutoModel.register(UNetConfig, UNetModel)
+
+__all__ = ["UNetConfig", "UNetModel"]

requirements.txt

@@ -1 +1,2 @@
-torch >= 2.7.1
+torch >= 2.7.1
+transformers >= 4.55.2

unet_config.py

@@ -0,0 +1,21 @@
+from transformers import PretrainedConfig
+
+
+class UNetConfig(PretrainedConfig):
+    model_type = "unet"
+
+    def __init__(
+        self,
+        in_channels=3,
+        out_channels=1,
+        pad=True,
+        bilinear=True,
+        normalization=None,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.pad = pad
+        self.bilinear = bilinear
+        self.normalization = normalization

unet_hf.py

@@ -0,0 +1,20 @@
+from transformers import PreTrainedModel
+from .unet import UNet
+from .unet_config import UNetConfig
+
+
+class UNetModel(PreTrainedModel):
+    config_class = UNetConfig
+
+    def __init__(self, config: UNetConfig):
+        super().__init__(config)
+        self.model = UNet(
+            in_channels=config.in_channels,
+            out_channels=config.out_channels,
+            pad=config.pad,
+            bilinear=config.bilinear,
+            normalization=config.normalization,
+        )
+
+    def forward(self, x):
+        return self.model(x)