Create app.py

app.py

@@ -0,0 +1,102 @@
+import requests
+import json
+import gradio as gr
+from huggingface_hub import InferenceClient
+from datetime import datetime
+
+class GaiaAgent:
+    def __init__(self, model_name="mistralai/Mixtral-8x7B-Instruct-v0.1"):
+        self.client = InferenceClient(model=model_name)
+        self.system_prompt = """You are solving GAIA benchmark questions. Follow these rules:
+        - Think step by step.
+        - End with: FINAL ANSWER: [answer].
+        - Answer must be short: a number, few words, or comma-separated list.
+        - No units ($, %) unless specified.
+        - No articles (a, an, the) in strings."""
+
+    def get_questions(self, split="validation"):
+        try:
+            url = f"https://huggingface.co/datasets/gaia-benchmark/GAIA/resolve/main/{split}/metadata.jsonl"
+            response = requests.get(url)
+            return [json.loads(line) for line in response.text.splitlines()]
+        except Exception as e:
+            print(f"Error fetching questions: {e}")
+            return []
+
+    def answer_question(self, question):
+        prompt = f"{self.system_prompt}\nQuestion: {question['question']}\nReasoning:"
+        response = self.client.text_generation(
+            prompt,
+            max_new_tokens=500,
+            temperature=0.1,
+            do_sample=False
+        )
+        return {
+            "task_id": question["task_id"],
+            "answer": self._extract_final_answer(response),
+            "reasoning": response
+        }
+
+    def _extract_final_answer(self, text):
+        if "FINAL ANSWER:" in text:
+            return text.split("FINAL ANSWER:")[-1].strip()
+        return text.strip()
+
+def run_evaluation(hf_username, full_name, question_limit=5):
+    agent = GaiaAgent()
+    questions = agent.get_questions()[:question_limit]
+    
+    if not questions:
+        return "Error: No se pudieron cargar las preguntas", ""
+    
+    results = []
+    for q in questions:
+        result = agent.answer_question(q)
+        results.append(f"**Pregunta {q['task_id']}:**\n{q['question']}\n\n**Respuesta:** {result['answer']}\n\n---")
+    
+    # Simular generación de certificado
+    certificate_info = f"""
+    ⭐ Certificado GAIA Benchmark ⭐
+    
+    Nombre: {full_name}
+    Usuario HF: {hf_username}
+    Fecha: {datetime.now().strftime('%Y-%m-%d')}
+    Preguntas respondidas: {len(results)}
+    
+    ¡Felicitaciones por completar el benchmark!
+    """
+    
+    return certificate_info, "\n".join(results)
+
+# Interfaz Gradio
+with gr.Blocks(title="Agente GAIA Benchmark") as demo:
+    gr.Markdown("""
+    # 🌟 Agente para GAIA Benchmark
+    Ejecuta evaluación y obtén tu certificado
+    """)
+    
+    with gr.Row():
+        with gr.Column():
+            hf_username = gr.Textbox(label="Usuario de Hugging Face", visible=False)
+            full_name = gr.Textbox(label="Nombre completo (aparecerá en el certificado)")
+            question_slider = gr.Slider(1, 20, value=5, label="Número de preguntas a evaluar")
+            submit_btn = gr.Button("Ejecutar Evaluación", variant="primary")
+        
+        with gr.Column():
+            certificate_output = gr.Textbox(label="Certificado", interactive=False)
+            answers_output = gr.Markdown(label="Respuestas Generadas")
+    
+    # Integración con login de HF
+    demo.load(
+        lambda: gr.Textbox(value=gr.Request().username if gr.Request() else ""),
+        outputs=hf_username
+    )
+    
+    submit_btn.click(
+        run_evaluation,
+        inputs=[hf_username, full_name, question_slider],
+        outputs=[certificate_output, answers_output]
+    )
+
+if __name__ == "__main__":
+    demo.launch()