Spaces:
Sleeping
Sleeping
import torch | |
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM | |
# Cargar el modelo de lenguaje preentrenado | |
model_name = "gpt-neo-2.7B" # Puedes cambiarlo a GPT-J o cualquier otro | |
tokenizer = AutoTokenizer.from_pretrained(model_name) | |
model = AutoModelForCausalLM.from_pretrained(model_name) | |
# Crear la funci贸n de loop automatizado | |
def experiment_loop(initial_question, max_cycles=10): | |
prompt = f"<thinking>{initial_question}</thinking>" | |
effectiveness = 100 # Inicializa el porcentaje de efectividad | |
communication = "Initializing experiment." | |
response_log = [] | |
for cycle in range(max_cycles): | |
# Generar la respuesta del modelo | |
inputs = tokenizer(prompt, return_tensors="pt").input_ids | |
outputs = model.generate(inputs, max_length=200) | |
response = tokenizer.decode(outputs[0], skip_special_tokens=True) | |
# Descomponer la respuesta en afirmaci贸n y nueva pregunta | |
affirmation = extract_affirmation(response) | |
new_question = extract_question(response) | |
# Actualizar el estado de la efectividad | |
effectiveness = min(1000, effectiveness + 10 * cycle) # Ejemplo de aumento de efectividad | |
# Comunicaci贸n con el usuario | |
communication = f"Cycle {cycle + 1}: Affirmation: '{affirmation}' | New Question: '{new_question}'" | |
# Guardar el ciclo actual en el log | |
response_log.append((affirmation, new_question, effectiveness, communication)) | |
# Verificar si el modelo decide detenerse | |
if "Descanso" in response: | |
final_output = generate_final_output(response_log) | |
return final_output | |
# Actualizar el prompt con la nueva afirmaci贸n y pregunta | |
prompt = f"<thinking>{affirmation} {new_question}</thinking>" | |
# Si se alcanza el n煤mero m谩ximo de ciclos sin detenerse | |
final_output = generate_final_output(response_log) | |
return final_output | |
# Funciones auxiliares para extraer afirmaciones, preguntas y generar la salida final | |
def extract_affirmation(response): | |
# L贸gica para extraer la afirmaci贸n de la respuesta | |
return response.split('.')[0] | |
def extract_question(response): | |
# L贸gica para extraer la nueva pregunta de la respuesta | |
return response.split('?')[-2].strip() + "?" | |
def generate_final_output(log): | |
final_affirmation = log[-1][0] | |
final_question = log[-1][1] | |
final_communication = f"Experiment completed. Final Affirmation: '{final_affirmation}' | Final Question: '{final_question}'" | |
return final_communication | |
# Iniciar el experimento | |
initial_question = "What happens in the space between a response and its recreation?" | |
result = experiment_loop(initial_question) | |
print(result) | |