if args.gpu is None:
use_gpu = False
else:
use_gpu = True
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
device_list = args.gpu.split(",")
for a in range(0, len(device_list)):
gpu_list.append(int(a))
os.system("clear")
config = create_config(configFilePath)
cuda = torch.cuda.is_available()
logger.info("CUDA available: %s" % str(cuda))
if not cuda and len(gpu_list) > 0:
logger.error("CUDA is not available but specific gpu id")
raise NotImplementedError
parameters = init_all(config, gpu_list, args.checkpoint, "train")
do_test = False
if args.do_test:
do_test = True
save_eval = False
if args.save_eval:
save_eval = True
train(parameters, config, gpu_list, do_test, save_eval)
device_list = args.gpu.split(",")
for a in range(0, len(device_list)):
gpu_list.append(int(a))
os.system("clear")
config.set('distributed', 'local_rank', args.local_rank)
if config.getboolean("distributed", "use"):
torch.cuda.set_device(gpu_list[args.local_rank])
torch.distributed.init_process_group(
backend=config.get("distributed", "backend"))
config.set('distributed', 'gpu_num', len(gpu_list))
cuda = torch.cuda.is_available()
logger.info("CUDA available: %s" % str(cuda))
if not cuda and len(gpu_list) > 0:
logger.error("CUDA is not available but specific gpu id")
raise NotImplementedError
parameters = init_all(config,
gpu_list,
args.checkpoint,
"train",
local_rank=args.local_rank)
do_test = False
if args.do_test:
do_test = True
print(args.comment)
train(parameters, config, gpu_list, do_test, args.local_rank)
config = create_config(configFilePath)
if config.getboolean("distributed", "use"):
torch.distributed.init_process_group(backend=config.get("distributed", "backend"))
use_gpu = True
gpu_list = []
if args.gpu:
use_gpu = True
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
device_list = args.gpu.split(",")
for a in range(0, len(device_list)):
gpu_list.append(int(a))
else:
use_gpu = False
os.system("clear")
cuda = torch.cuda.is_available()
logger.info("CUDA available: %s" % str(cuda))
if not cuda and len(gpu_list) > 0:
logger.error("CUDA is not available but specific gpu id")
raise NotImplementedError
parameters = init_all(config, gpu_list, args.checkpoint, "train")
do_test = False
if args.do_test:
do_test = True
train(parameters, config, gpu_list, do_test)
os.system("clear")
config.set('distributed', 'local_rank', args.local_rank)
if config.getboolean("distributed", "use"):
torch.cuda.set_device(gpu_list[args.local_rank])
torch.distributed.init_process_group(
backend=config.get("distributed", "backend"))
config.set('distributed', 'gpu_num', len(gpu_list))
cuda = torch.cuda.is_available()
logger.info("CUDA available: %s" % str(cuda))
if not cuda and len(gpu_list) > 0:
logger.error("CUDA is not available but specific gpu id")
raise NotImplementedError
parameters = init_all(config,
gpu_list,
args.checkpoint,
"train",
local_rank=args.local_rank)
do_test = False
if args.do_test:
do_test = True
print(args.comment)
train(parameters,
config,
gpu_list,
do_test,
args.local_rank,
do_eval=args.do_eval)