L1_Overview_of_Multimodality (first lab Jupyter Notebook) - Contrastive learning model don’t fit generating a very noise loss curve graph when in training mode (“train = True”).
Please, within the “Model Training” section remove or comment code “net = Network()” in the “train_model” function. See bellow:
Before:
def train_model(epoch_count=10):#
net = Network()
lrs = []
losses = []
After:
def train_model(epoch_count=10):#
# net = Network()
lrs = []
losses = []
To add support for GPU model training on MacOS with ARM processors (M1, M2,…) modify the code in the “Define the Training Configuration” section as follows:
net = Network()
device = 'cpu'
if torch.cuda.is_available():
device = torch.device('cuda:0')
elif torch.backends.mps.is_available():
device = torch.device("mps")
net = net.to(device)
device