def main(args):
if args.src_seq_length > args.max_position_embeddings:
args.max_position_embeddings = args.src_seq_length
if args.task.lower() in ['gigaword', 'blank']:
finetune(args,
train_valid_datasets_provider, {},
end_of_epoch_callback_provider=metrics_func_provider,
forward_step=seq2seq_forward_step)
else:
raise NotImplementedError(args.task)
def main(args):
if args.src_seq_length > args.max_position_embeddings:
args.max_position_embeddings = args.src_seq_length
if args.task.lower() in [
'cnn_dm', 'cnn_dm_original', 'gigaword', 'blank',
'squad_generation', 'xsum', 'squad', 'squad_v1', 'extraction',
'cmrc'
]:
finetune(args,
train_valid_datasets_provider, {},
end_of_epoch_callback_provider=metrics_func_provider,
forward_step=seq2seq_forward_step)
else:
raise NotImplementedError(args.task)
def main(args):
model_kwargs = {}
if args.task.lower(
) == 'wsc' and args.cloze_eval and not args.wsc_negative:
from tasks.language_model.finetune import lm_forward_step
finetune(args,
train_valid_datasets_provider,
model_kwargs,
end_of_epoch_callback_provider=metrics_func_provider,
forward_step=lm_forward_step)
else:
processor = PROCESSORS[args.task.lower()](args)
pvp = PVPS[args.task.lower()]
model_kwargs[
"model_type"] = "cloze" if args.cloze_eval else "classification"
model_kwargs["multi_token"] = pvp.is_multi_token
model_kwargs["num_labels"] = len(processor.get_labels())
finetune(args,
train_valid_datasets_provider,
model_kwargs,
end_of_epoch_callback_provider=metrics_func_provider)
def main(args):
model_kwargs = {}
processor = PROCESSORS[args.task.lower()](args)
pvp = PVPS[args.task.lower()](args,
None,
processor.get_labels(),
args.seq_length,
pattern_id=args.pattern_id,
is_multi_token=args.multi_token,
num_prompt_tokens=args.num_prompt_tokens)
if args.continuous_prompt:
model_kwargs["spell_length"] = pvp.spell_length
if args.task.lower(
) == 'wsc' and args.cloze_eval and not args.wsc_negative:
from tasks.language_model.finetune import lm_forward_step
finetune(args,
train_valid_datasets_provider,
model_kwargs,
end_of_epoch_callback_provider=metrics_func_provider,
forward_step=lm_forward_step)
else:
if args.cloze_eval:
multi_token = pvp.is_multi_token
else:
multi_token = args.task.lower() in MULTI_CHOICE_DATASETS
args.multi_token = multi_token
if not multi_token:
model_kwargs[
"model_type"] = "multiple_choice" if args.cloze_eval else "classification"
model_kwargs["multi_token"] = False
model_kwargs["num_labels"] = len(processor.get_labels())
else:
model_kwargs["model_type"] = "multiple_choice"
model_kwargs["multi_token"] = True
model_kwargs["num_labels"] = 1
finetune(args,
train_valid_datasets_provider,
model_kwargs,
end_of_epoch_callback_provider=metrics_func_provider)
def main(args):
"""Main program."""
finetune(args,
None, {},
end_of_epoch_callback_provider=metrics_func_provider,
forward_step=lm_forward_step)