Change model to Llama

app.py

@@ -11,9 +11,12 @@ from langchain.schema import Document
 load_dotenv()
 hftoken = os.getenv("HFTOKEN")
 
+prompt_systeme = os.getenv("PROMPT_SYSTEM")
+
+
 login(token=hftoken)
 
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=hftoken)
+client = InferenceClient(model="meta-llama/Llama-3.3-70B-Instruct", token=hftoken)
 embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
 
 vector_store = None
@@ -76,7 +79,7 @@ def chatbot_response(message, history, system_message, max_tokens, temperature,
         yield status
 
     context = query_faiss(message) if vector_store else None
-    messages = [{"role": "system", "content": system_message + " Réponds uniquement en français."}]
+    messages = [{"role": "system", "content": system_message}]
     
     for val in history:
         if val[0]: messages.append({"role": "user", "content": val[0]})
@@ -120,7 +123,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         with gr.Column():
             gr.Markdown("## ⚙️ Paramètres")
             with gr.Accordion(" Paramètres avancés", open=False):
-                system_message = gr.Textbox(value="Réponds de façon simple et claire.", label="Message système")
+                system_message = gr.Textbox(value=prompt_systeme, label="Message système")
                 max_tokens = gr.Slider(1, 2048, value=800, step=1, label="Max tokens")  
                 temperature = gr.Slider(0.1, 4.0, value=0.3, step=0.1, label="Température")
                 top_p = gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p")