Update to use custom blocks for TuneAVideo.

unet/config.json

@@ -12,10 +12,10 @@
   "center_input_sample": false,
   "cross_attention_dim": 768,
   "down_block_types": [
-    "CrossAttnDownBlock3D",
-    "CrossAttnDownBlock3D",
-    "CrossAttnDownBlock3D",
-    "DownBlock3D"
+    "CrossAttnDownBlockInflated3D",
+    "CrossAttnDownBlockInflated3D",
+    "CrossAttnDownBlockInflated3D",
+    "DownBlockInflated3D"
   ],
   "downsample_padding": 1,
   "in_channels": 4,
@@ -28,14 +28,12 @@
   "resnet_time_scale_shift": "default",
   "sample_size": 64,
   "up_block_types": [
-    "UpBlock3D",
-    "CrossAttnUpBlock3D",
-    "CrossAttnUpBlock3D",
-    "CrossAttnUpBlock3D"
+    "UpBlockInflated3D",
+    "CrossAttnUpBlockInflated3D",
+    "CrossAttnUpBlockInflated3D",
+    "CrossAttnUpBlockInflated3D"
   ],
   "upcast_attention": false,
   "use_linear_projection": false,
-  "use_temporal_transformer": false, 
-  "use_temporal_conv": false,
-  "sub_blocks_type": "3d"
+  "use_temporal_transformer": false,
 }