def build_model(hps, log):
model = WaveNODE(hps)
n_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('number of parameters:', n_params)
state = {}
state['n_params'] = n_params
log.write('%s\n' % json.dumps(state))
log.flush()
return model
def build_model(hps):
model = WaveNODE(hps)
print('number of parameters:',
sum(p.numel() for p in model.parameters() if p.requires_grad))
return model
return model, g_epoch, g_step
if __name__ == "__main__":
global global_step
global start_time
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
args = parse_args()
log_path, load_path = mkdir(args, test=True)
log_speed = get_logger(log_path, args.model_name, test_speed=True)
synth_loader = load_dataset(args)
hps = Hyperparameters(args)
model = build_model(hps)
model, global_epoch, global_step = load_checkpoint(args.load_step, model)
model = WaveNODE.remove_weightnorm(model)
model.to(device)
model.eval()
if args.tol_synth != args.tol:
from model import NODEBlock
print('change tolerance to {}'.format(args.tol_synth))
for block in model.blocks:
if isinstance(block, NODEBlock):
block.chains[2].test_atol = args.tol_synth
block.chains[2].test_rtol = args.tol_synth
with torch.no_grad():
synthesize(model, args.temp, args.tol_synth, log_speed)
log_speed.close()