set specific gpu_id for inference, but still use gpu 0

[Amoko]

First, I set gpu_id in file: tools/inference.py
deploy_cfg['gpu_id'] = '2'

Then, I check the gpu device in file: vedastr/runners/inference_runner.py

    def load_checkpoint(self, filename, map_location='default', strict=True):
        self.logger.info('Load checkpoint from {}'.format(filename))

        if map_location == 'default':
            if self.use_gpu:
                device_id = torch.cuda.current_device()
                print('recog true used gpu:', device_id)
                map_location = lambda storage, loc: storage.cuda(device_id)

But I got print like this.

recog true used gpu: 0