()
| 528 | |
| 529 | |
| 530 | def main(): |
| 531 | |
| 532 | args, config = parse_option() |
| 533 | # print(config.dump()) |
| 534 | |
| 535 | # Setup CUDA, GPU & distributed training |
| 536 | if args.local_rank == -1: |
| 537 | device = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
| 538 | args.n_gpu = torch.cuda.device_count() |
| 539 | else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs |
| 540 | torch.cuda.set_device(args.local_rank) |
| 541 | device = torch.device("cuda", args.local_rank) |
| 542 | torch.distributed.init_process_group(backend='nccl', |
| 543 | timeout=timedelta(minutes=60)) |
| 544 | args.n_gpu = 1 |
| 545 | args.device = device |
| 546 | |
| 547 | # Setup logging |
| 548 | logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s', |
| 549 | datefmt='%m/%d/%Y %H:%M:%S', |
| 550 | level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN) |
| 551 | logger.warning("Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s" % |
| 552 | (args.local_rank, args.device, args.n_gpu, bool(args.local_rank != -1), args.fp16)) |
| 553 | |
| 554 | # Set seed |
| 555 | set_seed(args) |
| 556 | |
| 557 | if args.engine: |
| 558 | validate_trt(args, config) |
| 559 | |
| 560 | # Calibration |
| 561 | if args.calib: |
| 562 | args, model = setup(args) |
| 563 | calib(args, config, model) |
| 564 | |
| 565 | # Quantization-Aware Training |
| 566 | if args.train: |
| 567 | # args, model = setup(args) |
| 568 | train(args, config) |
| 569 | |
| 570 | |
| 571 | if __name__ == "__main__": |
no test coverage detected