Update app.py

app.py

@@ -75,12 +75,12 @@ def respond(
     history: list[tuple[str, str]],
     preset,
     temperature,
-    mirostat_tau,
+    lp_start,
+    lp_decay,
     mirostat_eta,
+    mirostat_tau,
     frequency_penalty,
     presence_penalty,
-    lp_start,
-    lp_decay,
     max_tokens
 ):
     
@@ -104,7 +104,7 @@ def respond(
         mirostat_mode=1,
         mirostat_tau=mirostat_tau,
         mirostat_eta=mirostat_eta,
-        max_tokens=128,
+        max_tokens=max_tokens,
         frequency_penalty=frequency_penalty,
         presence_penalty=presence_penalty,
         logits_processor=lambda ids, logits: custom_lp_logits_processor(ids, logits, lp_start, lp_decay, len(convo))
@@ -129,14 +129,14 @@ demo = gr.ChatInterface(
     title="EliGPT v1.3",
     additional_inputs=[
         gr.Radio(presets.keys(), label="Preset", info="Gaslight the model into acting a certain way - WARNING, IF YOU CHANGE THIS WHILE THERE ARE MESSAGES IN THE CHAT, THE MODEL WILL BECOME VERY SLOW FOR YOU", value="none"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature", info="How chaotic should the model be?"),
-        gr.Slider(minimum=0.0, maximum=10.0, value=3.0, step=0.5, label="Mirostat tau", info="Basically, how many drugs should the model be on?"),
-        gr.Slider(minimum=0.0, maximum=1.0, value=0.1, step=0.01, label="Mirostat eta", info="I don't even know man"),
-        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=0.0, step=0.01, label="Presence penalty", info='"Use lots of diverse words"'),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.8, step=0.1, label="Temperature", info="How chaotic should the model be?"),
         gr.Slider(minimum=0, maximum=512, value=32, 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=1.02, step=0.01, label="Length penalty decay factor", info='How fast should the stop likelihood increase?'),
-        gr.Slider(minimum=1, maximum=1024, value=256, step=1, label="Max new tokens", info="How many words can the model generate?"),
+        gr.Slider(minimum=0.5, maximum=1.5, value=1.02, step=0.01, 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="Mirostat eta", info="How grammatical the model is or something"),
+        gr.Slider(minimum=0.0, maximum=10.0, value=3.0, step=0.5, label="Mirostat tau", info="Lower number keeps hallucinations to a minimum"),
+        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?"),
     ],
 )