Update app.py

app.py

@@ -7,7 +7,7 @@ os.environ['CURL_CA_BUNDLE'] = ''
 
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
-config = LinearMappingConfig()
+config = LinearMappingConfig(image_from_pretrained=False, text_from_pretrained=False)
 model = LinearMapping(config)
 model.load_state_dict(torch.load("pytorch_model.bin", map_location=device))
 processor = LinearMappingProcessor(config)
@@ -35,7 +35,11 @@ def generate_image_captions(image, text):
 
 article = "This demo is originated from this paper: [original paper](https://arxiv.org/abs/2209.15162)"
 description = """
-### Expand GPT2's language capabilities to vision with CLIP!
+### Expand GPT2's language capabilities to vision with CLIP! 
+### Tips:
+- When no image is provided, the model degrades to a vanilla GPT2-Large!
+- When no description is provided, the model automatically generates a caption for this image.
+- Try appending 'Answer:' after your question, the model is more likely to give desired outputs this way.
 """
 demo = gr.Interface(
     fn=generate_image_captions,
@@ -55,4 +59,4 @@ demo = gr.Interface(
     description=description
 )
 
-demo.launch()
+demo.launch(share=True)