Pārlūkot izejas kodu

test during training default to FP16

5.0
Glenn Jocher pirms 4 gadiem
vecāks
revīzija
a1748a8d6e
1 mainītis faili ar 5 papildinājumiem un 3 dzēšanām
  1. +5
    -3
      test.py

+ 5
- 3
test.py Parādīt failu

@@ -23,6 +23,7 @@ def test(data,
verbose=False):
# Initialize/load model and set device
if model is None:
training = False
device = torch_utils.select_device(opt.device, batch_size=batch_size)
half = device.type != 'cpu' # half precision only supported on CUDA

@@ -42,11 +43,12 @@ def test(data,
if device.type != 'cpu' and torch.cuda.device_count() > 1:
model = nn.DataParallel(model)

training = False
else: # called by train.py
device = next(model.parameters()).device # get model device
half = False
training = True
device = next(model.parameters()).device # get model device
half = device.type != 'cpu' # half precision only supported on CUDA
if half:
model.half() # to FP16

# Configure
model.eval()

Notiek ielāde…
Atcelt
Saglabāt