def setup_interact_args():
"""
Sets up the arguments for evaluation.
"""
parser = argparse.ArgumentParser()
group = parser.add_argument_group('interact')
group.add_argument('--method',
type=str,
default='nucleus',
help='Decoding method to use.')
group.add_argument('--cuda',
type=bool,
default=torch.cuda.is_available(),
help='Device for training.')
group.add_argument('--max_len',
type=int,
default=100,
help='Maximum length of the decoded sequence.')
group.add_argument('--top_p',
type=float,
default=0.9,
help='Top-p parameter for nucleus sampling.')
group.add_argument('--top_k',
type=int,
default=100,
help='Top-k parameter for topk sampling.')
setup_data_args(parser)
setup_model_args(parser)
return parser.parse_args()
def setup_train_args():
"""
Sets up the training arguments.
"""
parser = argparse.ArgumentParser()
group = parser.add_argument_group('train')
group.add_argument('--max_epochs',
type=int,
default=15,
help='Maximum number of epochs for training.')
group.add_argument('--cuda',
type=bool,
default=torch.cuda.is_available(),
help='Device for training.')
# TODO XLNet produces NaN with apex
group.add_argument('--mixed',
type=bool,
default=True,
help='Use mixed precision training.')
group.add_argument('--learning_rate',
type=float,
default=1e-4,
help='Learning rate for the model.')
group.add_argument('--batch_size',
type=int,
default=64,
help='Batch size during training.')
group.add_argument('--patience',
type=int,
default=5,
help='Number of patience epochs before termination.')
group.add_argument('--grad_accum_steps',
type=int,
default=2,
help='Number of steps for grad accum.')
group.add_argument('--eval_every_step',
type=int,
default=3000,
help='Evaluation frequency in steps.')
group.add_argument('--local_rank',
type=int,
default=-1,
help='Local rank for the script.')
setup_data_args(parser)
setup_model_args(parser)
return parser.parse_args()
def setup_eval_args():
"""
Sets up the arguments for interaction.
"""
parser = argparse.ArgumentParser()
parser.add_argument('--model_file',
type=str,
default=None,
help='Path to the file of the model.')
parser.add_argument('--ckpt_name',
type=str,
default='last',
choices=['last', 'best'],
help='Name of the checkpoint to load.')
parser.add_argument('--decoding',
type=str,
default='topk',
choices=list(METHODS),
help='Decoding method to use.')
parser.add_argument('--no_cuda',
action='store_true',
default=torch.cuda.is_available(),
help='Device for training.')
parser.add_argument('--top_p',
type=float,
default=0.9,
help='Top-p parameter for nucleus sampling.')
parser.add_argument('--top_k',
type=int,
default=100,
help='Top-k parameter for topk sampling.')
parser.add_argument('--min_len',
type=int,
default=0,
help='Minimum length of the response sentence.')
parser.add_argument('--seed',
type=int,
default=None,
help='Random seed for interactive mode.')
setup_data_args(parser)
setup_model_args(parser)
return parser.parse_args()
def setup_train_args():
"""
Sets up the training arguments.
"""
parser = argparse.ArgumentParser()
parser.add_argument('--config',
type=str,
default=None,
help='Path of the config file.')
parser.add_argument('--max_epochs',
type=int,
default=25,
help='Maximum number of epochs for training.')
parser.add_argument('--no_cuda',
action='store_true',
help='Device for training.')
# TODO XLNet produces NaN with apex
parser.add_argument('--fp16',
action='store_true',
help='Use fp16 precision training.')
parser.add_argument('--lr',
type=float,
default=1e-5,
help='Learning rate for the model.')
parser.add_argument('--batch_size',
type=int,
default=64,
help='Batch size during training.')
parser.add_argument('--patience',
type=int,
default=5,
help='Number of patience epochs before termination.')
parser.add_argument('--schedule',
type=str,
default='noam',
choices=['noam', 'noamwd'],
help='Type of learning rate scheduling.')
parser.add_argument('--warmup_steps',
type=int,
default=16000,
help='Number of warmup steps.')
parser.add_argument('--total_steps',
type=int,
default=1000000,
help='Number of optimization steps.')
parser.add_argument('--grad_accum_steps',
type=int,
default=2,
help='Number of steps for grad accum.')
parser.add_argument('--local_rank',
type=int,
default=-1,
help='Local rank for the script.')
parser.add_argument('--notebook',
action='store_true',
help='Set true if you are using IPython notebook.')
parser.add_argument('--clip_grad',
type=float,
default=None,
help='Gradient clipping norm value.')
parser.add_argument('--seed',
type=int,
default=None,
help='Random seed for training.')
setup_data_args(parser)
setup_model_args(parser)
return parser.parse_args()