Update app.py

app.py

@@ -86,34 +86,35 @@ def respond(
         response += token
         yield response
 
+ci = gr.ChatInterface(
+    respond,
+    additional_inputs_accordion=gr.Accordion(label="Options", open=True),
+    additional_inputs=[
+        gr.Radio(presets.keys(), label="Personality preset", info="VERY SLIGHTLY influence the model's personality [WARNING, IF YOU CHANGE THIS WHILE THERE ARE MESSAGES IN THE CHAT THE MODEL WILL BECOME VERY SLOW]", value="Default"),
+        # ("The model will become slow" is bc this uncaches the prompt and prompt processing is a big part of the generation time)
+        gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Min_p", info="Lower values give it more \"personality\""),
+        gr.Slider(minimum=0.1, maximum=4.0, value=1.5, step=0.1, label="Temperature", info="How chaotic should the model be?"),
+        gr.Slider(minimum=0, maximum=512, value=5, step=1, label="Length penalty start", info='When should the model start being more likely to shut up?'),
+        gr.Slider(minimum=0.5, maximum=1.5, value=1.015, step=0.001, label="Length penalty decay factor", info='How fast should that stop likelihood increase?'),
+        gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Frequency penalty", info='"Don\'repeat yourself"'),
+        gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Presence penalty", info='"Use lots of diverse words"'),
+        gr.Slider(minimum=1, maximum=1024, value=1024, step=1, label="Max new tokens", info="How many words can the model generate at most?"),
+    ],
+  )
+
 
 with gr.Blocks(css=".bubble-gap {gap: 6px !important}", theme="shivi/calm_seafoam") as demo:
   gr.Markdown("# EliGPT v1.3")
   gr.Markdown("Llama 3 8b finetuned on 2.5k of my discord messages. [Train your own clone!](https://gist.github.com/Elijah-Bodden/1964bd02fcd19efef65f6e0cd92881c4)")
   with gr.Accordion("Q&A:", open=False):
-      gr.Markdown("""Q Why is the model so fucking slow  
-        A The model might be slow if it hasn't run recently or a lot of people are using it (it's running on llama.cpp on a single a very slow cpu). You can duplicate the space to get your own (free) instance with no wait times.  
-        Q Why is the model so dumb  
-        A Llama 3 8b is impressive, but it's still tiny. This model is basically what you'd get if you shoved my brain into a toddler's head - it's just too small to be smart  
-        Q Either it just made something up or I don't know you at all  
-        A Probably the former. It's prone to hallucinating facts and opinions I don't hold. Take everything it says with a big grain of salt  
+      gr.Markdown("""Q: Why is the model so fucking slow  
+        A: The model might be slow if it hasn't run recently or a lot of people are using it (it's running on llama.cpp on a single a very slow cpu). You can duplicate the space to get your own (free) instance with no wait times.  
+        Q: Why is the model so dumb  
+        A: Llama 3 8b is impressive, but it's still tiny. This model is basically what you'd get if you shoved my brain into a toddler's head - it's just too small to be smart  
+        Q: Either it just made something up or I don't know you at all  
+        A: Probably the former. It's prone to hallucinating facts and opinions I don't hold. Take everything it says with a big grain of salt  
       """)
-  gr.ChatInterface(
-    respond,
-    additional_inputs_accordion=gr.Accordion(label="Options", open=True),
-    # additional_inputs=[
-    #     gr.Radio(presets.keys(), label="Personality preset", info="VERY SLIGHTLY influence the model's personality [WARNING, IF YOU CHANGE THIS WHILE THERE ARE MESSAGES IN THE CHAT THE MODEL WILL BECOME VERY SLOW]", value="Default"),
-    #     # ("The model will become slow" is bc this uncaches the prompt and prompt processing is a big part of the generation time)
-    #     gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Min_p", info="Lower values give it more \"personality\""),
-    #     gr.Slider(minimum=0.1, maximum=4.0, value=1.5, step=0.1, label="Temperature", info="How chaotic should the model be?"),
-    #     gr.Slider(minimum=0, maximum=512, value=5, step=1, label="Length penalty start", info='When should the model start being more likely to shut up?'),
-    #     gr.Slider(minimum=0.5, maximum=1.5, value=1.015, step=0.001, label="Length penalty decay factor", info='How fast should that stop likelihood increase?'),
-    #     gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Frequency penalty", info='"Don\'repeat yourself"'),
-    #     gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Presence penalty", info='"Use lots of diverse words"'),
-    #     gr.Slider(minimum=1, maximum=1024, value=1024, step=1, label="Max new tokens", info="How many words can the model generate at most?"),
-    # ],
-  )
-
+  ci.render()
 
 if __name__ == "__main__":
     demo.launch()