File size: 1,709 Bytes
0d16e19 a87db35 0d16e19 333ea63 05da434 333ea63 a87db35 4722f73 0d16e19 333ea63 05da434 333ea63 0d16e19 05da434 0d16e19 333ea63 0d16e19 333ea63 0d16e19 a87db35 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
class EndpointHandler:
def __init__(self, model_dir):
self.tokenizer = AutoTokenizer.from_pretrained(model_dir)
self.model = AutoModelForSeq2SeqLM.from_pretrained(model_dir)
self.model.eval()
def preprocess(self, data):
# Validar la entrada
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.")
# Tokenizar el texto de entrada
input_text = "Generate a valid JSON capturing data from this text: " + data["inputs"]
tokens = self.tokenizer(input_text, return_tensors="pt", truncation=True, padding="max_length", max_length=1000)
return tokens
def inference(self, tokens):
# Parámetros de generación
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
}
# Generar salida con el modelo
with torch.no_grad():
outputs = self.model.generate(**tokens, **generate_kwargs)
return outputs
def postprocess(self, outputs):
# Decodificar la salida generada
decoded_output = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
return {"response": decoded_output}
def __call__(self, data):
tokens = self.preprocess(data)
outputs = self.inference(tokens)
result = self.postprocess(outputs)
return result
|