Create config.json

config.json

@@ -0,0 +1,17 @@
+{
+  "model_type": "vision-encoder-decoder",
+  "vision_config": {
+    "hidden_size": 768,
+    "num_hidden_layers": 12,
+    "num_attention_heads": 12,
+    "image_size": 224,
+    "patch_size": 16
+  },
+  "text_config": {
+    "hidden_size": 768,
+    "num_hidden_layers": 12,
+    "num_attention_heads": 12,
+    "vocab_size": 32000,
+    "max_position_embeddings": 1024
+  }
+}