MCPcopy
hub / github.com/DingXiaoH/RepVGG / main

Function main

quantization/quant_qat_train.py:100–132  ·  view source on GitHub ↗
()

Source from the content-addressed store, hash-verified

98 return optimizer
99
100def main():
101 args = parser.parse_args()
102
103 if args.seed is not None:
104 random.seed(args.seed)
105 torch.manual_seed(args.seed)
106 cudnn.deterministic = True
107 warnings.warn('You have chosen to seed training. '
108 'This will turn on the CUDNN deterministic setting, '
109 'which can slow down your training considerably! '
110 'You may see unexpected behavior when restarting '
111 'from checkpoints.')
112
113 if args.gpu is not None:
114 warnings.warn('You have chosen a specific GPU. This will completely '
115 'disable data parallelism.')
116
117 if args.dist_url == "env://" and args.world_size == -1:
118 args.world_size = int(os.environ["WORLD_SIZE"])
119
120 args.distributed = args.world_size > 1 or args.multiprocessing_distributed
121
122 ngpus_per_node = torch.cuda.device_count()
123 if args.multiprocessing_distributed:
124 # Since we have ngpus_per_node processes per node, the total world_size
125 # needs to be adjusted accordingly
126 args.world_size = ngpus_per_node * args.world_size
127 # Use torch.multiprocessing.spawn to launch distributed processes: the
128 # main_worker process function
129 mp.spawn(main_worker, nprocs=ngpus_per_node, args=(ngpus_per_node, args))
130 else:
131 # Simply call main_worker function
132 main_worker(args.gpu, ngpus_per_node, args)
133
134
135

Callers 1

quant_qat_train.pyFile · 0.70

Calls 1

main_workerFunction · 0.85

Tested by

no test coverage detected