Upload Adapter

config.json

@@ -1,7 +1,12 @@
 {
+  "architectures": [
+    "Adapter"
+  ],
   "auto_map": {
-    "AutoConfig": "config.AdapterConfig"
+    "AutoConfig": "config.AdapterConfig",
+    "AutoModel": "model.Adapter"
   },
   "model_type": "archinetai/adapter-A-v1",
+  "torch_dtype": "float32",
   "transformers_version": "4.24.0"
 }

model.py

@@ -0,0 +1,61 @@
+import torch 
+from torch import Tensor, nn
+from transformers import PreTrainedModel
+from .config import AdapterConfig 
+
+
+class Model(nn.Module):
+
+    def __init__(
+        self,
+        num_channels: int,
+        num_filters: int,
+        window_length: int,
+        stride: int, 
+    ):
+        super().__init__()
+        self.stride = stride
+        padding = window_length // 2 - stride // 2
+        self.conv = nn.Conv1d(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=window_length,
+            stride=stride,
+            padding=padding,
+            padding_mode="reflect",
+            bias=False,
+        )
+        self.decode = nn.ConvTranspose1d(
+            in_channels=num_filters,
+            out_channels=num_channels,  
+            kernel_size=window_length,
+            stride=stride,
+            padding=padding,
+            bias=False,
+        )
+    
+    def encode(self, x: Tensor) -> Tensor:
+        return torch.tanh(self.conv(x))
+
+
+class Adapter(PreTrainedModel):
+
+    config_class = AdapterConfig
+
+    def __init__(self, config: AdapterConfig):
+        super().__init__(config)
+
+        self.model = Model(
+            num_channels=2,
+            num_filters=128,
+            window_length=128,
+            stride=64
+        )
+    
+    def encode(self, x):
+        return self.model.encode(x)
+
+    def decode(self, x):
+        return self.model.decode(x)
+    
+

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e428e6cb74fc8e479aed1bb50537e1be4fe06202815abd4562245d639d410185
+size 263143