Configure cache directories and add system prompt for local model

- Set specific cache directories for Transformers and HuggingFace libraries
- Add system prompt to guide model's language-adaptive behavior
- Modify call_model function to include system prompt in message generation
- Improve model initialization with explicit cache directory specification

app.py

@@ -10,6 +10,10 @@ import os
 from dotenv import load_dotenv
 load_dotenv()
 
+# Configurar directorio de caché en un lugar con permisos
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
+os.environ["HF_HOME"] = "/tmp/hf_home"
+
 # HuggingFace token
 HUGGINGFACE_TOKEN = os.environ.get("HUGGINGFACE_TOKEN", os.getenv("HUGGINGFACE_TOKEN"))
 print(f"Token HuggingFace: {HUGGINGFACE_TOKEN}")
@@ -21,12 +25,14 @@ MODEL_NAME = "HuggingFaceTB/SmolLM2-1.7B-Instruct"
 print(f"Loading model {MODEL_NAME} locally...")
 tokenizer = AutoTokenizer.from_pretrained(
     MODEL_NAME,
-    token=HUGGINGFACE_TOKEN  # Add token for authentication
+    token=HUGGINGFACE_TOKEN,  # Add token for authentication
+    cache_dir="/tmp/transformers_cache"  # Especificar directorio de caché
 )
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME, 
     device_map="auto",
-    token=HUGGINGFACE_TOKEN  # Add token for authentication
+    token=HUGGINGFACE_TOKEN,  # Add token for authentication
+    cache_dir="/tmp/transformers_cache"  # Especificar directorio de caché
 )
 
 # Create a pipeline to facilitate generation
@@ -52,8 +58,11 @@ def call_model(state: MessagesState):
     Returns:
         dict: A dictionary containing the generated text and the thread ID
     """
+    # System prompt to guide the model's behavior
+    system_prompt = "You are a friendly Chatbot. Always reply in the language in which the user is writing to you."
+    
     # Convert LangChain messages to a format that the local model can understand
-    prompt = ""
+    prompt = f"System: {system_prompt}\n\n"
     for msg in state["messages"]:
         if isinstance(msg, HumanMessage):
             prompt += f"User: {msg.content}\n"