Update app

app.py

@@ -5,6 +5,7 @@ import os
 from dotenv import load_dotenv
 from huggingface_hub import login
 from transformers import BitsAndBytesConfig
+from functools import lru_cache
 
 load_dotenv()
 
@@ -32,13 +33,19 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
-def generate_response(message, temperature=0.7, max_new_tokens=500):
+def generate_response(message, temperature=0.7, max_new_tokens=256):
     try:
         response = pipe(
             message,
             temperature=temperature,
             max_new_tokens=max_new_tokens,
-            do_sample=True
+            do_sample=True,
+            top_k=50,
+            top_p=0.9,
+            repetition_penalty=1.2,
+            pad_token_id=tokenizer.pad_token_id,
+            eos_token_id=tokenizer.eos_token_id,
+            batch_size=1
         )
         return response[0]['generated_text']
     except Exception as e:
@@ -50,7 +57,7 @@ demo = gr.Interface(
     inputs=[
         gr.Textbox(label="Votre message", placeholder="Entrez votre message ici..."),
         gr.Slider(minimum=0.1, maximum=1.0, value=0.7, label="Température"),
-        gr.Slider(minimum=10, maximum=2000, value=500, step=10, label="Nombre de tokens")
+        gr.Slider(minimum=10, maximum=3000, value=256, step=10, label="Nombre de tokens")
     ],
     outputs=gr.Textbox(label="Réponse"),
     title="Chat avec Sacha-Mistral",