PyTorch - NotImplementedError at model fit using MAC M1 MPS in Notebooks - macos

Question

I am getting error:
NotImplementedError: The operator 'aten::hardswish_' is not currently implemented for the MPS device, goes on to say, Comment https://github.com/pytorch/pytorch/issues/77764, but nothing there.
I have gone through the appropriate set up environment (conda activate 'en' and 'torch.backends.mps.is_available()' returns TRUE
Enviro:
Python Platform: macOS-13.2-arm64-arm-64bit
PyTorch Version: 1.13.1
Python 3.9.13 (main, Aug 25 2022, 18:24:45)
[Clang 12.0.0 ]
GPU is NOT AVAILABLE
MPS is AVAILABLE
Target device is mps
Googling indicates that some operators are not currently supported by PyTorch. If that's the case, should my implementation be different. Has anyone had experience in this, or will not work? (obviously work with cpu).
Getting the error here:
Input In [9], in <cell line: 4>()
13 imgs = list(img.to(device) for img in imgs)
14 annotations = [{k: v.to(device) for k, v in t.items()} for t in annotations]
---> 15 loss_dict = model(imgs, annotations)
16 losses = sum(loss for loss in loss_dict.values())
18 optimizer.zero_grad()
Setup:
def get_model_instance_segmentation(num_classes):
# load an instance segmentation model pre-trained pre-trained on COCO
#model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=False)
model = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_320_fpn(pretrained=False)
# get number of input features for the classifier
in_features = model.roi_heads.box_predictor.cls_score.in_features
# replace the pre-trained head with a new one
model.roi_heads.box_predictor = FastRCNNPredictor(in_features, num_classes)
return model
for img_, annotations in tr_loader:
list(img.to(device) for img in img_)
annotations = [{k: v.to(device) for k, v in t.items()} for t in annotations]
#print(annotations)
for img_, annotations in val_loader:
list(img.to(device) for img in img_)
annotations = [{k: v.to(device) for k, v in t.items()} for t in annotations]
#print(annotations)
model = get_model_instance_segmentation(config.num_classes)
if torch.backends.mps.is_available():
model.to(device)
print(device)
# parameters
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.SGD(
params, lr=config.lr, momentum=config.momentum, weight_decay=config.weight_decay
)
len_trloader = len(tr_loader)
len_valloader = len(val_loader)
train_loss = []
val_loss = []
min_valid_loss = np.inf
for epoch in range(config.num_epochs):
print(f"Epoch: {epoch}/{config.num_epochs}")
model.train()
i = 0
j = 0
for imgs, annotations in tr_loader:
i += 1
imgs = list(img.to(device) for img in imgs)
annotations = [{k: v.to(device) for k, v in t.items()} for t in annotations]
loss_dict = model(imgs, annotations)
losses = sum(loss for loss in loss_dict.values())
optimizer.zero_grad()
losses.backward()
optimizer.step()
train_loss.append(losses.tolist())
print(f"Iteration_Train: {i}/{len_trloader}, Loss: {losses}")
valid_loss = 0.0
for imgs, annotations in val_loader:
j += 1
imgs = list(img.to(device) for img in imgs)
annotations = [{k: v.to(device) for k, v in t.items()} for t in annotations]
loss_dict_val = model(imgs, annotations)
losses_val = sum(loss_val for loss_val in loss_dict_val.values())
val_loss.append(losses_val.tolist())
print(f"Iteration_Val: {i}/{len_valloader}, Loss: {losses_val}")