def build_MTLtrainer(args, device_id, model, optims, optims_inner, loss):
"""Builds the trainer class for MTL
Args:
device_id (int): the GPU id to be used
model (models.model_builder.MTLAbsSummarizer')
optims (list[models.optimizers.Optimizer])
optims_inner (list[models.optimizer.Optimizer])
loss (models.loss.NMTLossCompute)
Returns:
A object in type pipelines.trainer_abs_mtl.MTLTrainer.
"""
device = "cpu" if args.visible_gpus == '-1' else "cuda"
grad_accum_count = args.accum_count
n_gpu = args.world_size
if device_id >= 0:
gpu_rank = int(args.gpu_ranks[device_id])
else:
gpu_rank = 0
n_gpu = 0
print('gpu_rank %d' % gpu_rank)
# Prepare tensorborad writer
writer = SummaryWriter(args.log_path, comment="Unmt")
# Prepare report manager
report_manager = ReportMgr(args.report_every,
start_time=-1,
tensorboard_writer=writer)
report_inner_manager = ReportMgr(args.report_inner_every,
start_time=-1,
tensorboard_writer=writer)
# Prepare trainer
trainer = MTLTrainer(args, model, optims, optims_inner, loss,
grad_accum_count, n_gpu, gpu_rank, report_manager,
report_inner_manager)
# Show # of (trainable) parameters
if (model):
n_params = _tally_parameters(model)
trainable_n_params = _tally_trainable_parameters(model)
logger.info('Number of parameters: %d' % n_params)
logger.info('Number of trainalbe parameters: %d' % trainable_n_params)
return trainer
def build_trainer(args, device_id, model, optims, tokenizer):
"""
Simplify `Trainer` creation based on user `opt`s*
Args:
opt (:obj:`Namespace`): user options (usually from argument parsing)
model (:obj:`onmt.models.NMTModel`): the model to train
fields (dict): dict of fields
optim (:obj:`onmt.utils.Optimizer`): optimizer used during training
data_type (str): string describing the type of data
e.g. "text", "img", "audio"
model_saver(:obj:`onmt.models.ModelSaverBase`): the utility object
used to save the model
"""
device = "cpu" if args.visible_gpus == '-1' else "cuda"
grad_accum_count = args.accum_count
n_gpu = args.world_size
if device_id >= 0:
gpu_rank = int(args.gpu_ranks[device_id])
else:
gpu_rank = 0
n_gpu = 0
print('gpu_rank %d' % gpu_rank)
tensorboard_log_dir = args.model_path
writer = SummaryWriter(tensorboard_log_dir, comment="Unmt")
report_manager = ReportMgr(args.report_every,
start_time=-1,
tensorboard_writer=writer)
symbols = {
'BOS': tokenizer.vocab['[unused1]'],
'EOS': tokenizer.vocab['[unused2]'],
'PAD': tokenizer.vocab['[PAD]'],
'SEG': tokenizer.vocab['[unused3]'],
'UNK': tokenizer.vocab['[UNK]']
}
gen_loss = abs_loss(args,
model.generator,
symbols,
tokenizer.vocab,
device,
train=True)
pn_loss = CrossEntropyLossCompute().to(device)
trainer = Trainer(args, model, optims, tokenizer, gen_loss, pn_loss,
grad_accum_count, n_gpu, gpu_rank, report_manager)
# print(tr)
if (model):
n_params = _tally_parameters(model)
logger.info('* number of parameters: %d' % n_params)
return trainer
def build_trainer(args, model, optim):
gpu_rank = 0
print('gpu_rank %d' % gpu_rank)
report_manager = ReportMgr(args.report_every, start_time=-1)
trainer = Trainer(args, model, optim, report_manager)
return trainer
def build_trainer(args, device_id, model, optims, loss):
"""
Simplify `Trainer` creation based on user `opt`s*
Args:
opt (:obj:`Namespace`): user options (usually from argument parsing)
model (:obj:`onmt.models.NMTModel`): the model to train
fields (dict): dict of fields
optim (:obj:`onmt.utils.Optimizer`): optimizer used during training
data_type (str): string describing the type of data
e.g. "text", "img", "audio"
model_saver(:obj:`onmt.models.ModelSaverBase`): the utility object
used to save the model
"""
device = "cpu" if args.visible_gpus == '-1' else "cuda"
grad_accum_count = args.accum_count
n_gpu = args.world_size
if device_id >= 0:
gpu_rank = int(args.gpu_ranks[device_id])
else:
gpu_rank = 0
n_gpu = 0
print('gpu_rank %d' % gpu_rank)
report_manager = None
if not args.debug:
tensorboard_log_dir = args.model_path
writer = SummaryWriter(tensorboard_log_dir, comment="Unmt")
report_manager = ReportMgr(args.report_every,
start_time=-1,
tensorboard_writer=writer)
trainer = Trainer(args, model, optims, loss, grad_accum_count, n_gpu,
gpu_rank, report_manager)
# print(tr)
if (model):
n_params = _tally_parameters(model)
logger.info('* number of parameters: %d' % n_params)
return trainer
def build_trainer(args, device_id, model, optims,loss):
device = "cpu" if args.visible_gpus == '-1' else "cuda"
grad_accum_count = args.accum_count
n_gpu = args.world_size
if device_id >= 0:
gpu_rank = int(args.gpu_ranks[device_id])
else:
gpu_rank = 0
n_gpu = 0
print('gpu_rank %d' % gpu_rank)
tensorboard_log_dir = args.model_path
writer = SummaryWriter(tensorboard_log_dir, comment="Unmt")
report_manager = ReportMgr(args.report_every, start_time=-1, tensorboard_writer=writer)
trainer = Trainer(args, model, optims, loss, grad_accum_count, n_gpu, gpu_rank, report_manager)
if (model):
n_params = _tally_parameters(model)
logger.info('* number of parameters: %d' % n_params)
return trainer
def build_trainer(args, device_id, model,
optim):
grad_accum_count = args.accum_count
n_gpu = 1
if device_id < 0:
n_gpu = 0
gpu_rank = 0
tensorboard_log_dir = args.model_path
writer = SummaryWriter(tensorboard_log_dir, comment="Unmt")
report_manager = ReportMgr(args.report_every, start_time=-1, tensorboard_writer=writer)
trainer = Trainer(args, model, optim, grad_accum_count, n_gpu, gpu_rank, report_manager)
n_params, enc, dec = _tally_parameters(model)
logger.info('encoder: %d' % enc)
logger.info('decoder: %d' % dec)
logger.info('* number of parameters: %d' % n_params)
return trainer
