Add the upscaling UNet

model_index.json

@@ -9,12 +9,20 @@
     "diffusers",
     "CLIPTextModel"
   ],
+  "text_unet": [
+    "diffusers",
+    "GLIDETextToImageUNetModel"
+  ],
   "tokenizer": [
     "transformers",
     "GPT2Tokenizer"
   ],
-  "unet": [
+  "upscale_noise_scheduler": [
+    "diffusers",
+    "GlideDDIMScheduler"
+  ],
+  "upscale_unet": [
     "diffusers",
-    "UNetGLIDEModel"
+    "GLIDESuperResUNetModel"
   ]
 }

{unet → text_unet}/config.json

@@ -1,5 +1,5 @@
 {
-  "_class_name": "UNetGLIDEModel",
+  "_class_name": "GLIDETextToImageUNetModel",
   "attention_resolutions": [
     2,
     4,
@@ -22,7 +22,6 @@
   "num_res_blocks": 3,
   "out_channels": 6,
   "resblock_updown": true,
-  "transformer_dim": 512,
   "use_checkpoint": false,
   "use_fp16": false,
   "use_scale_shift_norm": true

{unet → text_unet}/diffusion_model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0f8313e755b8aea63d67ae7177e073e901c9812136511df4759feea84f840b9
+oid sha256:f2c5b7116a5a45e0c67113e53c7d1204ed7e54dd301894c130d7d5820c184c34
 size 1235135533

upscale_noise_scheduler/scheduler_config.json

@@ -0,0 +1,5 @@
+{
+  "_class_name": "GlideDDIMScheduler",
+  "beta_schedule": "linear",
+  "timesteps": 1000
+}

upscale_unet/config.json

@@ -0,0 +1,30 @@
+{
+  "_class_name": "GLIDESuperResUNetModel",
+  "attention_resolutions": [
+    8,
+    16,
+    32
+  ],
+  "channel_mult": [
+    1,
+    1,
+    2,
+    2,
+    4,
+    4
+  ],
+  "conv_resample": true,
+  "dims": 2,
+  "dropout": 0.1,
+  "in_channels": 6,
+  "model_channels": 192,
+  "num_head_channels": 64,
+  "num_heads": 1,
+  "num_heads_upsample": 1,
+  "num_res_blocks": 2,
+  "out_channels": 6,
+  "resblock_updown": true,
+  "use_checkpoint": false,
+  "use_fp16": false,
+  "use_scale_shift_norm": true
+}

upscale_unet/diffusion_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2a378df148fc34f0a30228cfb9f3c4457819d2e4293bdd3841ca65a9eea79c4
+size 1244328341