Update app.py

app.py

@@ -170,15 +170,20 @@ with gr.Blocks(analytics_enabled=False, css=custom_css) as demo:
         with gr.Column():
             gr.Markdown(
                 """
-                # Demo platform for 🥐 CroissantLLMChat
+                ## Demo platform for 🥐 CroissantLLMChat
                 
                 The model is of small size (1.3B), about 130 times smaller than GPT3.
-                As such, it logically exhibits reduced understanding, reasoning and knowlege capacities. 
-                For industrial uses, we recommend finetuning the model, but trained a Chat version to allow for experimenting and showcase the capabilities for it's size.
-                We recommend testing it for open-ended writing tasks, tips, translations, etc...
-                The model can hallucinate and generate incorrect or even toxic content.
+                As such, it's generalist Chat version logically exhibits reduced understanding, reasoning and knowledge capacities. 
+                For industrial uses, we recommend finetuning the model, but trained this Chat version to allow for experimenting and to showcase the capabilities for it's size.
+                
+                ## Usage recommendations 
+                We recommend testing the chat model for open-ended writing tasks, tips, translations, etc... 
+                We find direct instructions to work best, and performance to drop after the first round of interactions.
+                We limit the length of the conversation so clear the Chat between tests !
 
+                ## Errors
                 The demo is linked to an endpoint that auto-shutdowns after 15mn. If error message appears, wait about 5 minutes and test again once the server is back up !
+                The model can hallucinate and generate incorrect or even toxic content.
                 """
             )