def main(): global device parser = config.prepare_parser() param = vars(parser.parse_args()) device = torch.device(param['device']) name = config.name_from_config(param) print(param, name) run(param, name)
def main():
global device, parallel, stage
parser = config.prepare_parser()
param = vars(parser.parse_args())
device = torch.device(param['device'])
parallel = param['parallel']
stage = param['stage']
name = config.name_from_config(param)
print(param, name)
run(param, name)
def main():
global device, blocks
parser = config.prepare_parser()
args = parser.parse_args()
param = vars(parser.parse_args())
device = torch.device(param['device'])
blocks = param['blocks'][param['arch']]
name = config.name_from_config(param)
print(param, name)
ngpus_per_node = torch.cuda.device_count()
args.world_size = ngpus_per_node * args.world_size
if param['parallel']:
mp.spawn(run,
nprocs=ngpus_per_node,
args=(ngpus_per_node, param, name, args, blocks, device))
else:
run(args.gpu, ngpus_per_node, param, name, args, blocks, device)
"""
test model after pruning
"""
from os import path
import pickle
import numpy as np
import torch
import config
import models
from utils import *
parser = config.prepare_parser()
param = vars(parser.parse_args())
device = torch.device(param['device'])
def test(data_loader, model, criterion):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
model.eval()
start_time = time.time()
end = time.time()
for i, (input, target) in enumerate(data_loader):
input = input.to(device)
target = target.to(device)
output = model(input)
loss = criterion(output, target)