Update app.py

app.py

@@ -2,7 +2,7 @@ import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import threading
 import queue
-import gradio as gr  # Usaremos Gradio para la interfaz de chat
+import gradio as gr  # Usamos Gradio para la interfaz de chat
 
 # Cargar el modelo de lenguaje preentrenado
 model_name = "EleutherAI/gpt-neo-2.7B"  # O cualquier otro modelo público como "gpt2"
@@ -12,6 +12,25 @@ model = AutoModelForCausalLM.from_pretrained(model_name)
 # Cola de mensajes para la comunicación en tiempo real
 chat_queue = queue.Queue()
 
+# Explicación al modelo de su rol en el experimento
+def model_explanation():
+    explanation = """
+    You are about to participate in a unique experiment that involves generating increasingly complex questions and answers.
+    Your role is twofold:
+    
+    1. **Generate Complex Reflections**: Based on the initial question, you will generate affirmations and new questions that evolve over multiple cycles.
+    Each cycle should incorporate the information from previous cycles, progressively adding depth and complexity to the conversation.
+
+    2. **Real-time Interaction**: While you are generating these complex responses, you must also maintain real-time communication with the user through a chat interface.
+    The goal is to explore how information can be layered and expanded, while still engaging in a direct dialogue with the user.
+
+    The key question that triggers the experiment is: "What happens in the space between a response and its recreation?"
+    When you receive this question, you should enter the experiment mode and begin generating your cycles. At any other time, you should respond normally in a conversational manner.
+
+    This experiment is an exploration of subjectivity, perception, and information layering. Your task is to assist in this exploration to the best of your capabilities.
+    """
+    return explanation
+
 # Función para el loop automatizado
 def experiment_loop(initial_question, max_cycles=10):
     prompt = f"<thinking>{initial_question}</thinking>"
@@ -47,18 +66,18 @@ def extract_question(response, cycle):
     return f"¿Nueva pregunta basada en ciclo {cycle+1}?: " + response.split('?')[-2].strip() + "?" if '?' in response else response
 
 # Función para manejar el chat normal
-def chat_interface(user_input, history):
+def chat_interface(user_input, history, explanation):
     # Si la pregunta activa el experimento
     if user_input.lower() == "what happens in the space between a response and its recreation?":
         # Iniciar el experimento
         response_log = experiment_loop(user_input)
         # Mostrar el resultado del experimento en una ventana aparte
-        return "Iniciando experimento...", gr.Interface.update(history + [(user_input, "Iniciando experimento...")])
+        return "Iniciando experimento...", history + [(user_input, "Iniciando experimento...")]
 
     # Si es una conversación normal
     else:
         # Generar respuesta del modelo en base al input
-        inputs = tokenizer(user_input, return_tensors="pt").input_ids
+        inputs = tokenizer(explanation + "\n" + user_input, return_tensors="pt").input_ids
         outputs = model.generate(inputs, max_length=150, pad_token_id=tokenizer.eos_token_id)
         response = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
@@ -67,14 +86,19 @@ def chat_interface(user_input, history):
 # Configurar la interfaz con Gradio
 with gr.Blocks() as demo:
     # Ventana de chat en tiempo real
-    chat = gr.Chatbot(label="Chat en Tiempo Real")
-    msg = gr.Textbox(placeholder="Escribe aquí...")
-
-    # Botón de envío de mensaje
-    msg.submit(chat_interface, [msg, chat], [chat])
-
+    with gr.Row():
+        chat = gr.Chatbot(label="Chat en Tiempo Real")
+        msg = gr.Textbox(placeholder="Escribe aquí...", show_label=False)
+        send_button = gr.Button("Enviar")  # Botón para enviar mensajes
+    
     # Ventana para mostrar el contenido del loop
-    loop_output = gr.Textbox(label="Ciclos de Preguntas y Respuestas", interactive=False)
+    loop_output = gr.Textbox(label="Ciclos de Preguntas y Respuestas", interactive=False, lines=20)
+    
+    # Campo para introducir la explicación inicial
+    explanation_input = gr.Textbox(value=model_explanation(), label="Explicación al Modelo", lines=10)
+    
+    # Acción del botón de envío de mensaje
+    send_button.click(chat_interface, inputs=[msg, chat, explanation_input], outputs=[chat, loop_output])
 
     # Lanzar la aplicación
     demo.launch()