is_training=False,
num_gpus=args.num_gpus,
)
# Create callbacks for train and eval modes
train_callback = SimpleLossLoggerCallback(
tensors=train_tensors,
print_func=lambda x: logging.info(str(round(x[0].item(), 3))),
tb_writer=nf.tb_writer,
get_tb_values=lambda x: [["loss", x[0]]],
step_freq=train_steps_per_epoch,
)
eval_callback = nemo.core.EvaluatorCallback(
eval_tensors=eval_tensors,
user_iter_callback=lambda x, y: eval_iter_callback(x, y),
user_epochs_done_callback=lambda x: eval_epochs_done_callback(
x,
intents_label_ids=data_desc.intents_label_ids,
slots_label_ids=data_desc.slots_label_ids,
graph_fold=f'{nf.work_dir}/graphs',
normalize_cm=True,
),
tb_writer=nf.tb_writer,
eval_step=train_steps_per_epoch,
)
# Create callback to save checkpoints
ckpt_callback = CheckpointCallback(folder=nf.checkpoint_dir,
epoch_freq=args.save_epoch_freq,
step_freq=args.save_step_freq)
is_training=False,
num_gpus=args.num_gpus,
)
# Create callbacks for train and eval modes
train_callback = SimpleLossLoggerCallback(
tensors=train_tensors,
print_func=lambda x: logging.info(str(round(x[0].item(), 3))),
tb_writer=nf.tb_writer,
get_tb_values=lambda x: [["loss", x[0]]],
step_freq=train_steps_per_epoch,
)
eval_callback = nemo.core.EvaluatorCallback(
eval_tensors=eval_tensors,
user_iter_callback=lambda x, y: eval_iter_callback(x, y, eval_data_layer),
user_epochs_done_callback=lambda x: eval_epochs_done_callback(
x, f'{nf.work_dir}/graphs'),
tb_writer=nf.tb_writer,
eval_step=train_steps_per_epoch,
)
# Create callback to save checkpoints
ckpt_callback = CheckpointCallback(folder=nf.checkpoint_dir,
epoch_freq=args.save_epoch_freq,
step_freq=args.save_step_freq)
lr_policy_fn = get_lr_policy(args.lr_policy,
total_steps=args.num_epochs *
train_steps_per_epoch,
warmup_ratio=args.lr_warmup_proportion)
