else:
raise NotImplementedError("Currently, the %s model is not supported" %
(train_config["model"]))
model.load_state_dict(
torch.load(os.path.join(finetune_config["load_model_dir"],
"epoch%d.pt" % (best_epochs["dev"])),
map_location=torch.device("cpu")))
model.increase_net(finetune_config)
if not all([
train_config[key] == finetune_config[key] for key in [
"max_npv", "max_npe", "max_npvl", "max_npel", "max_ngv",
"max_nge", "max_ngvl", "max_ngel", "share_emb", "share_arch"
]
]):
model.increase_input_size(finetune_config)
if not all([
train_config[key] == finetune_config[key] for key in [
"predict_net", "predict_net_hidden_dim",
"predict_net_num_heads", "predict_net_mem_len",
"predict_net_mem_init", "predict_net_recurrent_steps"
]
]):
new_predict_net = model.create_predict_net(
finetune_config["predict_net"],
pattern_dim=model.predict_net.pattern_dim,
graph_dim=model.predict_net.graph_dim,
hidden_dim=finetune_config["predict_net_hidden_dim"],
num_heads=finetune_config["predict_net_num_heads"],
recurrent_steps=finetune_config["predict_net_recurrent_steps"],
mem_len=finetune_config["predict_net_mem_len"],
