Update app.py

app.py

@@ -14,6 +14,7 @@ def generate(
     temperature=0.4,
     top_p=0.95,
     top_k=50,
+    repetition_penalty=1.1,
     max_new_tokens=256,
 ):
     pipe = load_model(model_name)
@@ -24,7 +25,7 @@ def generate(
     else:
         prompt = f"<|im_start|>user\n{user_input}<|im_end|>\n<|im_start|>assistant\n"
     outputs = pipe(prompt, max_new_tokens=max_new_tokens, do_sample=True,
-                   temperature=temperature, top_k=top_k, top_p=top_p, repetition_penalty=1.10)
+                   temperature=temperature, top_k=top_k, top_p=top_p, repetition_penalty=repetition_penalty)
     return outputs[0]["generated_text"]
 
 model_choices = ["Locutusque/Hyperion-3.0-Mixtral-3x7B", "M4-ai/NeuralReyna-Mini-1.8B-v0.2", "Locutusque/Hyperion-3.0-Mistral-7B-DPO", "Locutusque/Hyperion-3.0-Mistral-7B-alpha", "M4-ai/tau-1.8B", "Locutusque/Hercules-4.0-Mistral-v0.2-7B", "Locutusque/Hercules-2.5-Mistral-7B", "M4-ai/tau-0.5B"]
@@ -35,6 +36,7 @@ g = gr.Interface(
         gr.components.Dropdown(choices=model_choices, label="Model", value=model_choices[0], interactive=True),
         gr.components.Textbox(lines=2, label="Prompt", value="Write me a Python program that calculates the factorial of a given number."),
         gr.components.Slider(minimum=0, maximum=1, value=0.4, label="Temperature"),
+        gr.components.Slider(minimum=1, maximum=2, value=1.1, label="Repetition Penalty"),
         gr.components.Slider(minimum=0, maximum=1, value=0.95, label="Top p"),
         gr.components.Slider(minimum=0, maximum=100, step=1, value=50, label="Top k"),
         gr.components.Slider(minimum=1, maximum=2048, step=1, value=1024, label="Max tokens"),