Update config for diffusers AutoencoderRAE refactor

config.json

@@ -5,9 +5,7 @@
   "decoder_intermediate_size": 4096,
   "decoder_num_attention_heads": 16,
   "decoder_num_hidden_layers": 28,
-  "encoder_cls": "dinov2",
   "encoder_input_size": 224,
-  "encoder_name_or_path": "facebook/dinov2-with-registers-base",
   "image_size": null,
   "latents_mean": null,
   "latents_std": [
@@ -222737,5 +222735,18 @@
   "patch_size": 16,
   "reshape_to_2d": true,
   "scaling_factor": 1.0,
-  "use_encoder_loss": false
-}
+  "use_encoder_loss": false,
+  "encoder_type": "dinov2",
+  "encoder_hidden_size": 768,
+  "encoder_patch_size": 14,
+  "encoder_norm_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "encoder_norm_std": [
+    0.229,
+    0.224,
+    0.225
+  ]
+}

conversion_metadata.json

@@ -1,12 +0,0 @@
-{
-  "source": "/data/boyang/models",
-  "encoder_cls": "dinov2",
-  "encoder_name_or_path": "facebook/dinov2-with-registers-base",
-  "decoder_checkpoint": "decoders/dinov2/wReg_base/ViTXL_n08/model.pt",
-  "stats_checkpoint": "stats/dinov2/wReg_base/imagenet1k/stat.pt",
-  "variant": "ViTXL_n08",
-  "dataset_name": "imagenet1k",
-  "decoder_config_name": "ViTXL",
-  "missing_decoder_keys": [],
-  "unexpected_decoder_keys": []
-}