AI/question_answering.py

# Author: Ricardo Lisboa Santos
# Creation date: 2024-01-10

import torch
import torch_directml
from transformers import pipeline

def getDevice(DEVICE):
    device = None
    if DEVICE == "cpu":
        device = torch.device("cpu")
        dtype = torch.float32
    elif DEVICE == "cuda":
        device = torch.device("cuda")
        dtype = torch.float16
    elif DEVICE == "directml":
        device = torch_directml.device()
        dtype = torch.float16
    return device

def loadGenerator(device):
    generator = pipeline("question-answering") # .to(device)
    return generator

def query(generator, question, context):
    output = generator(
        question=question,
        context=context,
    )
    return output

def clearCache(DEVICE, generator):
    generator.tokenizer.save_pretrained("cache")
    generator.model.save_pretrained("cache")
    del generator
    if DEVICE == "directml":
        torch_directml.empty_cache()