handler.py

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
import json
import jsonschema
import re

class EndpointHandler:
    def __init__(self, model_dir):
        self.tokenizer = AutoTokenizer.from_pretrained(model_dir)
        self.model = AutoModelForSeq2SeqLM.from_pretrained(model_dir)
        self.model.eval()

        # Esquema de validación del JSON
        self.json_schema = {
            "type": "object",
            "properties": {
                "values": {
                    "type": "array",
                    "items": {
                        "type": "object",
                        "properties": {
                            "id": {"type": "string"},
                            "value": {"type": ["string", "array"]}
                        },
                        "required": ["id", "value"]
                    },
                },
            },
            "required": ["values"],
        }

    def preprocess(self, data):
        if not isinstance(data, dict) or "inputs" not in data or data["inputs"] is None:
            raise ValueError("La entrada debe ser un diccionario con la clave 'inputs' y un valor válido.")

        input_text = f"""
        Por favor, genera un JSON válido basado en las siguientes especificaciones:
        ... (Especificaciones del formato JSON omitidas por brevedad)
        Procesa el siguiente texto: "{data['inputs']}"
        """
        tokens = self.tokenizer(input_text, return_tensors="pt", truncation=True, padding="max_length", max_length=1000)
        return tokens

    def inference(self, tokens):
        generate_kwargs = {
            "max_length": 1000,
            "num_beams": 5,
            "do_sample": True,
            "temperature": 0.3,
            "top_k": 50,
            "top_p": 0.9,
            "repetition_penalty": 2.5
        }
        with torch.no_grad():
            outputs = self.model.generate(**tokens, **generate_kwargs)
        return outputs

    def clean_output(self, output):
        json_match = re.search(r"{.*}", output, re.DOTALL)
        if json_match:
            return json_match.group(0)
        return output

    def validate_json(self, decoded_output):
        cleaned_output = self.clean_output(decoded_output)
        try:
            json_data = json.loads(cleaned_output)
            jsonschema.validate(instance=json_data, schema=self.json_schema)
            return {"is_valid": True, "json_data": json_data}
        except json.JSONDecodeError as e:
            return {"is_valid": False, "error": f"Error decodificando JSON: {str(e)}", "raw_output": cleaned_output}
        except jsonschema.exceptions.ValidationError as e:
            return {"is_valid": False, "error": f"Error validando JSON: {str(e)}", "raw_output": cleaned_output}

    def postprocess(self, outputs):
        decoded_output = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        validation_result = self.validate_json(decoded_output)
        
        # Siempre imprimir la salida generada
        print(f"Texto generado: {decoded_output}")

        if not validation_result["is_valid"]:
            print(f"Error en la validación: {validation_result['error']}")
            print(f"Salida sin procesar: {validation_result.get('raw_output', 'No disponible')}")
            raise ValueError(f"JSON inválido: {validation_result['error']}")
        
        return {"response": validation_result["json_data"]}

    def __call__(self, data):
        tokens = self.preprocess(data)
        outputs = self.inference(tokens)
        result = self.postprocess(outputs)
        return result