Create config.json

config.json

@@ -0,0 +1,10 @@
+{
+  "model_type": "vision_language",
+  "vision_encoder": "ViT",
+  "text_encoder": "Transformer",
+  "hidden_size": 896,
+  "num_hidden_layers": 10,
+  "num_attention_heads": 14,
+  "image_size": 224,
+  "vocab_size": 32000
+}