def load_model(load_path: dict, device: torch.device) -> FullModel:
"""
Instantiate the model from saved dictionary of pretrained model.
Args:
num_graph_layers: Number of layers to use in the model.
num_classes: Number of classes in the dataset.
pretrain_load_path: Use pretrained weights.
Returns:
The instantiated model with the requested parameters.
"""
ckpt = torch.load(f=load_path, map_location=device)
model_params_dict = ckpt["model_params_dict"]
model_params_dict['graph_layers_sizes'] = model_params_dict[
'graph_layers_sizes'][1:]
model = FullModel(
model_type=model_params_dict['model_type'],
num_classes=model_params_dict['num_classes'],
num_node_features=model_params_dict['num_node_features'],
graph_layers_sizes=model_params_dict['graph_layers_sizes'],
text_embed_dim=model_params_dict['text_embed_dim'],
text_output_dim=model_params_dict['text_output_dim'],
linear_layers_sizes=model_params_dict['linear_layers_sizes'],
dropout_rate=model_params_dict['dropout_rate'],
vocab_size=model_params_dict['vocab_size'])
model.load_state_dict(state_dict=ckpt["model_state_dict"])
# transfer model to cpu or gpu
model = model.to(device=device)
return model, model_params_dict
### Create model
CRNN_model = CRNN(IN_SIZE, HIDDEN_SIZE, KERNEL_SIZE, STRIDE,
GRU_NUM_LAYERS)
attn_layer = AttnMech(HIDDEN_SIZE * NUM_DIRS)
apply_attn = ApplyAttn(HIDDEN_SIZE * 2, NUM_CLASSES)
### Download ready models
# checkpoint = torch.load('crnn_final', map_location=device)
# CRNN_model.load_state_dict(checkpoint['model_state_dict'])
# checkpoint = torch.load('attn_final', map_location=device)
# attn_layer.load_state_dict(checkpoint['model_state_dict'])
# checkpoint = torch.load('apply_attn_final', map_location=device)
# apply_attn.load_state_dict(checkpoint['model_state_dict'])
full_model = FullModel(CRNN_model, attn_layer, apply_attn)
print(full_model.to(device))
print(count_parameters(full_model))
#wandb.init()
#wandb.watch(full_model)
### Create optimizer
opt = torch.optim.Adam(full_model.parameters(), weight_decay=1e-5)
### Train_val loop
for n in range(NUM_EPOCHS):
train_epoch(full_model,
opt,
train_loader,
melspec_train,
GRU_NUM_LAYERS,
HIDDEN_SIZE,