def main():
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument("--data_dir", default=None, type=str, required=True,
help="The input data dir. Should contain the training files for the NER/POS task.")
parser.add_argument("--model_type", default=None, type=str, required=True,
help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()))
parser.add_argument("--model_name_or_path", default=None, type=str, required=True,
help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS))
parser.add_argument("--output_dir", default=None, type=str, required=True,
help="The output directory where the model predictions and checkpoints will be written.")
## Other parameters
parser.add_argument("--labels", default="", type=str,
help="Path to a file containing all labels. If not specified, NER/POS labels are used.")
parser.add_argument("--config_name", default="", type=str,
help="Pretrained config name or path if not the same as model_name")
parser.add_argument("--tokenizer_name", default="", type=str,
help="Pretrained tokenizer name or path if not the same as model_name")
parser.add_argument("--cache_dir", default=None, type=str,
help="Where do you want to store the pre-trained models downloaded from s3")
parser.add_argument("--max_seq_length", default=128, type=int,
help="The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded.")
parser.add_argument("--do_train", action="store_true",
help="Whether to run training.")
parser.add_argument("--do_eval", action="store_true",
help="Whether to run eval on the dev set.")
parser.add_argument("--do_predict", action="store_true",
help="Whether to run predictions on the test set.")
parser.add_argument("--do_predict_dev", action="store_true",
help="Whether to run predictions on the dev set.")
parser.add_argument("--do_predict_train", action="store_true")
parser.add_argument("--init_checkpoint", default=None, type=str,
help="initial checkpoint for train/predict")
parser.add_argument("--evaluate_during_training", action="store_true",
help="Whether to run evaluation during training at each logging step.")
parser.add_argument("--do_lower_case", action="store_true",
help="Set this flag if you are using an uncased model.")
parser.add_argument("--few_shot", default=-1, type=int,
help="num of few-shot exampes")
parser.add_argument("--per_gpu_train_batch_size", default=8, type=int,
help="Batch size per GPU/CPU for training.")
parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int,
help="Batch size per GPU/CPU for evaluation.")
parser.add_argument("--gradient_accumulation_steps", type=int, default=1,
help="Number of updates steps to accumulate before performing a backward/update pass.")
parser.add_argument("--learning_rate", default=5e-5, type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--weight_decay", default=0.0, type=float,
help="Weight decay if we apply some.")
parser.add_argument("--adam_epsilon", default=1e-8, type=float,
help="Epsilon for Adam optimizer.")
parser.add_argument("--max_grad_norm", default=1.0, type=float,
help="Max gradient norm.")
parser.add_argument("--num_train_epochs", default=3.0, type=float,
help="Total number of training epochs to perform.")
parser.add_argument("--max_steps", default=-1, type=int,
help="If > 0: set total number of training steps to perform. Override num_train_epochs.")
parser.add_argument("--warmup_steps", default=0, type=int,
help="Linear warmup over warmup_steps.")
parser.add_argument("--logging_steps", type=int, default=50,
help="Log every X updates steps.")
parser.add_argument("--save_steps", type=int, default=50,
help="Save checkpoint every X updates steps.")
parser.add_argument("--save_only_best_checkpoint", action="store_true",
help="Save only the best checkpoint during training")
parser.add_argument("--eval_all_checkpoints", action="store_true",
help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number")
parser.add_argument("--no_cuda", action="store_true",
help="Avoid using CUDA when available")
parser.add_argument("--overwrite_output_dir", action="store_true",
help="Overwrite the content of the output directory")
parser.add_argument("--overwrite_cache", action="store_true",
help="Overwrite the cached training and evaluation sets")
parser.add_argument("--seed", type=int, default=42,
help="random seed for initialization")
parser.add_argument("--fp16", action="store_true",
help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit")
parser.add_argument("--fp16_opt_level", type=str, default="O1",
help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
"See details at https://nvidia.github.io/apex/amp.html")
parser.add_argument("--local_rank", type=int, default=-1,
help="For distributed training: local_rank")
parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.")
parser.add_argument("--server_port", type=str, default="", help="For distant debugging.")
parser.add_argument("--predict_langs", type=str, default="en", help="prediction languages")
parser.add_argument("--train_langs", default="en", type=str,
help="The languages in the training sets.")
parser.add_argument("--log_file", type=str, default=None, help="log file")
parser.add_argument("--eval_patience", type=int, default=-1, help="wait N times of decreasing dev score before early stop during training")
## SDE parameters
parser.add_argument("--max_ngram_size", default=10, type=int,
help="ngram size for each word")
parser.add_argument("--bpe_segment", type=int, default=1, help="whether to segment by BPE or by word")
parser.add_argument("--sde_latent", type=int, default=5000, help="sde latent emb size")
parser.add_argument("--use_sde_embed", action="store_true")
parser.add_argument("--add_sde_embed", action="store_true")
parser.add_argument("--tau", type=float, default=-1, help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--attention_t", type=float, default=1, help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--mlm_weight", type=float, default=-1, help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--mlm_lang", type=str, default='ur', help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--mlm_start_epoch", type=int, default=0, help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--mlm_end_epoch", type=int, default=0, help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--update_pretrained_epoch", type=int, default=0, help="wait N times of decreasing dev score before early stop during training")
parser.add_argument("--bpe_dropout", default=0, type=float)
parser.add_argument("--resample_dataset", default=0, type=float, help="set to 1 if resample at each epoch")
parser.add_argument("--fix_class", action='store_true')
# RecAdam parameters
parser.add_argument("--optimizer", type=str, default="RecAdam", choices=["Adam", "RecAdam"],
help="Choose the optimizer to use. Default RecAdam.")
parser.add_argument("--recadam_anneal_fun", type=str, default='sigmoid', choices=["sigmoid", "linear", 'constant'],
help="the type of annealing function in RecAdam. Default sigmoid")
parser.add_argument("--recadam_anneal_k", type=float, default=0.5, help="k for the annealing function in RecAdam.")
parser.add_argument("--recadam_anneal_t0", type=int, default=250, help="t0 for the annealing function in RecAdam.")
parser.add_argument("--recadam_anneal_w", type=float, default=1.0,
help="Weight for the annealing function in RecAdam. Default 1.0.")
parser.add_argument("--recadam_pretrain_cof", type=float, default=5000.0,
help="Coefficient of the quadratic penalty in RecAdam. Default 5000.0.")
parser.add_argument("--logging_Euclid_dist", action="store_true",
help="Whether to log the Euclidean distance between the pretrained model and fine-tuning model")
parser.add_argument("--start_from_pretrain", action="store_true",
help="Whether to initialize the model with pretrained parameters")
parser.add_argument("--albert_dropout", default=0.0, type=float,
help="The dropout rate for the ALBERT model")
parser.add_argument("--few_shot_extra_langs", type=str, default=None)
parser.add_argument("--few_shot_extra_langs_size", type=str, default=None)
args = parser.parse_args()
if os.path.exists(args.output_dir) and os.listdir(
args.output_dir) and args.do_train and not args.overwrite_output_dir:
raise ValueError(
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(
args.output_dir))
# Setup distant debugging if needed
if args.server_ip and args.server_port:
import ptvsd
print("Waiting for debugger attach")
ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True)
ptvsd.wait_for_attach()
# Setup CUDA, GPU & distributed training
if args.local_rank == -1 or args.no_cuda:
device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu")
args.n_gpu = torch.cuda.device_count()
else:
# Initializes the distributed backend which sychronizes nodes/GPUs
torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank)
torch.distributed.init_process_group(backend="nccl")
args.n_gpu = 1
args.device = device
# Setup logging
logging.basicConfig(handlers = [logging.FileHandler(args.log_file), logging.StreamHandler()],
format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt = '%m/%d/%Y %H:%M:%S',
level = logging.INFO if args.local_rank in [-1, 0] else logging.WARN)
logging.info("Input args: %r" % args)
logger.warning("Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16)
# Set seed
set_seed(args)
# Prepare NER/POS task
labels = get_labels(args.labels)
num_labels = len(labels)
# Use cross entropy ignore index as padding label id
# so that only real label ids contribute to the loss later
pad_token_label_id = CrossEntropyLoss().ignore_index
# Load pretrained model and tokenizer
# Make sure only the first process in distributed training loads model/vocab
if args.local_rank not in [-1, 0]:
torch.distributed.barrier()
args.model_type = args.model_type.lower()
if args.mlm_weight > 0:
#args.model_type = args.model_type+"_mlm"
config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type+"_mlm"]
else:
config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]
config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path,
num_labels=num_labels,
use_sde_embed=args.use_sde_embed,
add_sde_embed=args.add_sde_embed,
sde_latent=args.sde_latent,
mlm_weight=args.mlm_weight,
attention_t=args.attention_t,
fix_class=args.fix_class,
cache_dir=args.cache_dir if args.cache_dir else None)
tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=args.cache_dir if args.cache_dir else None)
if args.optimizer == 'RecAdam':
pretrained_model = model_class.from_pretrained(
args.model_name_or_path,
from_tf=bool(".ckpt" in args.model_name_or_path),
config=config,
cache_dir=args.cache_dir if args.cache_dir else None,
)
pretrained_model.to(args.device)
else:
pretrained_model = None
if args.init_checkpoint:
logger.info("loading from init_checkpoint={}".format(args.init_checkpoint))
model = model_class.from_pretrained(args.init_checkpoint,
config=config,
cache_dir=args.init_checkpoint)
else:
logger.info("loading from cached model = {}".format(args.model_name_or_path))
model = model_class.from_pretrained(args.model_name_or_path,
from_tf=bool(".ckpt" in args.model_name_or_path),
config=config,
cache_dir=args.cache_dir if args.cache_dir else None)
lang2id = config.lang2id if args.model_type == "xlm" else None
logger.info("Using lang2id = {}".format(lang2id))
# Make sure only the first process in distributed training loads model/vocab
if args.local_rank == 0:
torch.distributed.barrier()
model.to(args.device)
logger.info("Training/evaluation parameters %s", args)
# Training
if args.do_train:
train_dataset = load_and_cache_examples(args, tokenizer, labels, pad_token_label_id, mode="train", lang=args.train_langs, lang2id=lang2id, few_shot=args.few_shot)
global_step, tr_loss = train(args, train_dataset, model, tokenizer, labels, pad_token_label_id, lang2id, pretrained_model=pretrained_model)
logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)
# Saving best-practices: if you use default names for the model,
# you can reload it using from_pretrained()
if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
# Create output directory if needed
if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
os.makedirs(args.output_dir)
# Save model, configuration and tokenizer using `save_pretrained()`.
# They can then be reloaded using `from_pretrained()`
# Take care of distributed/parallel training
logger.info("Saving model checkpoint to %s", args.output_dir)
model_to_save = model.module if hasattr(model, "module") else model
model_to_save.save_pretrained(args.output_dir)
tokenizer.save_pretrained(args.output_dir)
# Good practice: save your training arguments together with the model
torch.save(args, os.path.join(args.output_dir, "training_args.bin"))
# Initialization for evaluation
results = {}
if args.init_checkpoint:
best_checkpoint = args.init_checkpoint
elif os.path.exists(os.path.join(args.output_dir, 'checkpoint-best')):
best_checkpoint = os.path.join(args.output_dir, 'checkpoint-best')
else:
best_checkpoint = args.output_dir
best_f1 = 0
# Evaluation
if args.do_eval and args.local_rank in [-1, 0]:
tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
checkpoints = [args.output_dir]
if args.eval_all_checkpoints:
checkpoints = list(os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)))
logging.getLogger("pytorch_transformers.modeling_utils").setLevel(logging.WARN)
logger.info("Evaluate the following checkpoints: %s", checkpoints)
for checkpoint in checkpoints:
global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else ""
model = model_class.from_pretrained(checkpoint)
model.to(args.device)
result, _ = evaluate(args, model, tokenizer, labels, pad_token_label_id, mode="dev", prefix=global_step, lang=args.train_langs, lang2id=lang2id)
if result["f1"] > best_f1:
best_checkpoint = checkpoint
best_f1 = result["f1"]
if global_step:
result = {"{}_{}".format(global_step, k): v for k, v in result.items()}
results.update(result)
output_eval_file = os.path.join(args.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
for key in sorted(results.keys()):
writer.write("{} = {}\n".format(key, str(results[key])))
writer.write("best checkpoint = {}, best f1 = {}\n".format(best_checkpoint, best_f1))
# Prediction
if args.do_predict and args.local_rank in [-1, 0]:
logger.info("Loading the best checkpoint from {}\n".format(best_checkpoint))
tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
model = model_class.from_pretrained(best_checkpoint)
model.to(args.device)
output_test_results_file = os.path.join(args.output_dir, "test_results.txt")
with open(output_test_results_file, "a") as result_writer:
for lang in args.predict_langs.split(','):
if not os.path.exists(os.path.join(args.data_dir, lang, 'test.{}'.format(args.model_name_or_path))):
logger.info("Language {} does not exist".format(lang))
continue
result, predictions = evaluate(args, model, tokenizer, labels, pad_token_label_id, mode="test", lang=lang, lang2id=lang2id)
# Save results
result_writer.write("=====================\nlanguage={}\n".format(lang))
for key in sorted(result.keys()):
result_writer.write("{} = {}\n".format(key, str(result[key])))
# Save predictions
output_test_predictions_file = os.path.join(args.output_dir, "test_{}_predictions.txt".format(lang))
infile = os.path.join(args.data_dir, lang, "test.{}".format(args.model_name_or_path))
idxfile = infile + '.idx'
save_predictions(args, predictions, output_test_predictions_file, infile, idxfile)
if args.do_predict_train and args.local_rank in [-1, 0]:
logger.info("Loading the best checkpoint from {}\n".format(best_checkpoint))
tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
model = model_class.from_pretrained(best_checkpoint)
model.to(args.device)
output_test_results_file = os.path.join(args.output_dir, "test_results.txt")
with open(output_test_results_file, "a") as result_writer:
for lang in args.predict_langs.split(','):
if not os.path.exists(os.path.join(args.data_dir, lang, 'train.{}'.format(args.model_name_or_path))):
logger.info("Language {} does not exist".format(lang))
continue
result, predictions = evaluate(args, model, tokenizer, labels, pad_token_label_id, mode="train", lang=lang, lang2id=lang2id)
# Save results
result_writer.write("=====================\nlanguage={}\n".format(lang))
for key in sorted(result.keys()):
result_writer.write("{} = {}\n".format(key, str(result[key])))
# Save predictions
output_test_predictions_file = os.path.join(args.output_dir, "train_{}_predictions.txt".format(lang))
infile = os.path.join(args.data_dir, lang, "train.{}".format(args.model_name_or_path))
idxfile = infile + '.idx'
save_predictions(args, predictions, output_test_predictions_file, infile, idxfile, output_word_prediction=True)
# Predict dev set
if args.do_predict_dev and args.local_rank in [-1, 0]:
logger.info("Loading the best checkpoint from {}\n".format(best_checkpoint))
tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
model = model_class.from_pretrained(best_checkpoint)
model.to(args.device)
output_test_results_file = os.path.join(args.output_dir, "dev_results.txt")
with open(output_test_results_file, "w") as result_writer:
for lang in args.predict_langs.split(','):
if not os.path.exists(os.path.join(args.data_dir, lang, 'dev.{}'.format(args.model_name_or_path))):
logger.info("Language {} does not exist".format(lang))
continue
result, predictions = evaluate(args, model, tokenizer, labels, pad_token_label_id, mode="dev", lang=lang, lang2id=lang2id)
# Save results
result_writer.write("=====================\nlanguage={}\n".format(lang))
for key in sorted(result.keys()):
result_writer.write("{} = {}\n".format(key, str(result[key])))
# Save predictions
output_test_predictions_file = os.path.join(args.output_dir, "dev_{}_predictions.txt".format(lang))
infile = os.path.join(args.data_dir, lang, "dev.{}".format(args.model_name_or_path))
idxfile = infile + '.idx'
save_predictions(args, predictions, output_test_predictions_file, infile, idxfile)
def main():
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument(
"--data_dir",
default=None,
type=str,
required=True,
help=
"The input data dir. Should contain the training files for the NER/POS task."
)
parser.add_argument("--model_type",
default=None,
type=str,
required=True,
help="Model type selected in the list: " +
", ".join(MODEL_CLASSES.keys()))
parser.add_argument(
"--model_name_or_path",
default=None,
type=str,
required=True,
help="Path to pre-trained model or shortcut name selected in the list: "
+ ", ".join(ALL_MODELS))
parser.add_argument(
"--output_dir",
default=None,
type=str,
required=True,
help=
"The output directory where the model predictions and checkpoints will be written."
)
## Other parameters
parser.add_argument(
"--labels",
default="",
type=str,
help=
"Path to a file containing all labels. If not specified, NER/POS labels are used."
)
parser.add_argument(
"--config_name",
default="",
type=str,
help="Pretrained config name or path if not the same as model_name")
parser.add_argument(
"--tokenizer_name",
default="",
type=str,
help="Pretrained tokenizer name or path if not the same as model_name")
parser.add_argument(
"--cache_dir",
default=None,
type=str,
help=
"Where do you want to store the pre-trained models downloaded from s3")
parser.add_argument(
"--max_seq_length",
default=128,
type=int,
help=
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded.")
parser.add_argument("--do_train",
action="store_true",
help="Whether to run training.")
parser.add_argument("--do_eval",
action="store_true",
help="Whether to run eval on the dev set.")
parser.add_argument("--do_predict",
action="store_true",
help="Whether to run predictions on the test set.")
parser.add_argument("--do_predict_dev",
action="store_true",
help="Whether to run predictions on the dev set.")
parser.add_argument("--init_checkpoint",
default=None,
type=str,
help="initial checkpoint for train/predict")
parser.add_argument(
"--evaluate_during_training",
action="store_true",
help="Whether to run evaluation during training at each logging step.")
parser.add_argument(
"--do_lower_case",
action="store_true",
help="Set this flag if you are using an uncased model.")
parser.add_argument("--few_shot",
default=-1,
type=int,
help="num of few-shot exampes")
parser.add_argument("--per_gpu_train_batch_size",
default=8,
type=int,
help="Batch size per GPU/CPU for training.")
parser.add_argument("--per_gpu_eval_batch_size",
default=8,
type=int,
help="Batch size per GPU/CPU for evaluation.")
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help=
"Number of updates steps to accumulate before performing a backward/update pass."
)
parser.add_argument("--learning_rate",
default=5e-5,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--weight_decay",
default=0.0,
type=float,
help="Weight decay if we apply some.")
parser.add_argument("--adam_epsilon",
default=1e-8,
type=float,
help="Epsilon for Adam optimizer.")
parser.add_argument("--max_grad_norm",
default=1.0,
type=float,
help="Max gradient norm.")
parser.add_argument("--num_train_epochs",
default=3.0,
type=float,
help="Total number of training epochs to perform.")
parser.add_argument(
"--max_steps",
default=-1,
type=int,
help=
"If > 0: set total number of training steps to perform. Override num_train_epochs."
)
parser.add_argument("--warmup_steps",
default=0,
type=int,
help="Linear warmup over warmup_steps.")
parser.add_argument("--logging_steps",
type=int,
default=50,
help="Log every X updates steps.")
parser.add_argument("--save_steps",
type=int,
default=50,
help="Save checkpoint every X updates steps.")
parser.add_argument("--save_only_best_checkpoint",
action="store_true",
help="Save only the best checkpoint during training")
parser.add_argument(
"--eval_all_checkpoints",
action="store_true",
help=
"Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number"
)
parser.add_argument("--no_cuda",
action="store_true",
help="Avoid using CUDA when available")
parser.add_argument("--overwrite_output_dir",
action="store_true",
help="Overwrite the content of the output directory")
parser.add_argument(
"--overwrite_cache",
action="store_true",
help="Overwrite the cached training and evaluation sets")
parser.add_argument("--seed",
type=int,
default=42,
help="random seed for initialization")
parser.add_argument(
"--fp16",
action="store_true",
help=
"Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit"
)
parser.add_argument(
"--fp16_opt_level",
type=str,
default="O1",
help=
"For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
"See details at https://nvidia.github.io/apex/amp.html")
parser.add_argument("--local_rank",
type=int,
default=-1,
help="For distributed training: local_rank")
parser.add_argument("--server_ip",
type=str,
default="",
help="For distant debugging.")
parser.add_argument("--server_port",
type=str,
default="",
help="For distant debugging.")
parser.add_argument("--predict_langs",
type=str,
default="en",
help="prediction languages")
parser.add_argument("--train_langs",
default="en",
type=str,
help="The languages in the training sets.")
parser.add_argument("--log_file", type=str, default=None, help="log file")
parser.add_argument(
"--eval_patience",
type=int,
default=-1,
help=
"wait N times of decreasing dev score before early stop during training"
)
args = parser.parse_args()
if os.path.exists(args.output_dir) and os.listdir(
args.output_dir
) and args.do_train and not args.overwrite_output_dir:
raise ValueError(
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome."
.format(args.output_dir))
# Setup distant debugging if needed
if args.server_ip and args.server_port:
import ptvsd
print("Waiting for debugger attach")
ptvsd.enable_attach(address=(args.server_ip, args.server_port),
redirect_output=True)
ptvsd.wait_for_attach()
# Setup CUDA, GPU & distributed training
if args.local_rank == -1 or args.no_cuda:
device = torch.device("cuda" if torch.cuda.is_available()
and not args.no_cuda else "cpu")
args.n_gpu = torch.cuda.device_count()
else:
# Initializes the distributed backend which sychronizes nodes/GPUs
torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank)
torch.distributed.init_process_group(backend="nccl")
args.n_gpu = 1
args.device = device
# Setup logging
logging.basicConfig(
handlers=[logging.FileHandler(args.log_file),
logging.StreamHandler()],
format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S',
level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN)
logging.info("Input args: %r" % args)
logger.warning(
"Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
args.local_rank, device, args.n_gpu, bool(args.local_rank != -1),
args.fp16)
# Set seed
set_seed(args)
# Prepare NER/POS task
labels = get_labels(args.labels)
num_labels = len(labels)
# Use cross entropy ignore index as padding label id
# so that only real label ids contribute to the loss later
pad_token_label_id = CrossEntropyLoss().ignore_index
# Load pretrained model and tokenizer
# Make sure only the first process in distributed training loads model/vocab
if args.local_rank not in [-1, 0]:
torch.distributed.barrier()
args.model_type = args.model_type.lower()
config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]
# config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path,
# num_labels=num_labels,
# cache_dir=args.cache_dir if args.cache_dir else None)
config_path = "/blob/kaiyuan-result/token-level/thisistheother"
config = config_class.from_pretrained(
config_path,
cache_dir=args.cache_dir if args.cache_dir else None,
)
tokenizer = tokenizer_class.from_pretrained(
args.tokenizer_name
if args.tokenizer_name else args.model_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=args.cache_dir if args.cache_dir else None)
# if args.init_checkpoint:
# logger.info("loading from init_checkpoint={}".format(args.init_checkpoint))
# model = model_class.from_pretrained(args.init_checkpoint,
# config=config,
# cache_dir=args.init_checkpoint)
# else:
# logger.info("loading from cached model = {}".format(args.model_name_or_path))
# model = model_class.from_pretrained(args.model_name_or_path,
# from_tf=bool(".ckpt" in args.model_name_or_path),
# config=config,
# cache_dir=args.cache_dir if args.cache_dir else None)
model = XLMForTokenClassification(config)
model.from_pretrained(
model_path="/blob/kaiyuan-result/token-level/kaiyuan-1230-bfpos-80.pth",
config=config,
)
lang2id = config.lang2id if args.model_type == "xlm" else None
logger.info("Using lang2id = {}".format(lang2id))
# Make sure only the first process in distributed training loads model/vocab
if args.local_rank == 0:
torch.distributed.barrier()
model.to(args.device)
logger.info("Training/evaluation parameters %s", args)
# Training
if args.do_train:
train_dataset = load_and_cache_examples(args,
tokenizer,
labels,
pad_token_label_id,
mode="train",
lang=args.train_langs,
lang2id=lang2id,
few_shot=args.few_shot)
global_step, tr_loss = train(args, train_dataset, model, tokenizer,
labels, pad_token_label_id, lang2id)
logger.info(" global_step = %s, average loss = %s", global_step,
tr_loss)
# Saving best-practices: if you use default names for the model,
# you can reload it using from_pretrained()
if args.do_train and (args.local_rank == -1
or torch.distributed.get_rank() == 0):
# Create output directory if needed
if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
os.makedirs(args.output_dir)
# Save model, configuration and tokenizer using `save_pretrained()`.
# They can then be reloaded using `from_pretrained()`
# Take care of distributed/parallel training
# logger.info("Saving model checkpoint to %s", args.output_dir)
# torch.save(model.state_dict(), 'save.pt')
tokenizer.save_pretrained(args.output_dir)
# Good practice: save your training arguments together with the model
# torch.save(args, os.path.join(args.output_dir, "training_args.bin"))
# Initialization for evaluation
results = {}
if args.init_checkpoint:
best_checkpoint = args.init_checkpoint
elif os.path.exists(os.path.join(args.output_dir, 'checkpoint-best.pth')):
best_checkpoint = os.path.join(args.output_dir, 'checkpoint-best.pth')
else:
best_checkpoint = args.output_dir
best_f1 = 0
# Evaluation
if args.do_eval and args.local_rank in [-1, 0]:
tokenizer = tokenizer_class.from_pretrained(
args.output_dir, do_lower_case=args.do_lower_case)
model.load_state_dict(torch.load(best_checkpoint))
model.to(args.device)
result, _ = evaluate(args,
model,
tokenizer,
labels,
pad_token_label_id,
mode="dev",
prefix=global_step,
lang=args.train_langs,
lang2id=lang2id)
results.update(result)
output_eval_file = os.path.join(args.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
for key in sorted(results.keys()):
writer.write("{} = {}\n".format(key, str(results[key])))
writer.write("best checkpoint = {}, best f1 = {}\n".format(
best_checkpoint, best_f1))
# Prediction
if args.do_predict and args.local_rank in [-1, 0]:
logger.info(
"Loading the best checkpoint from {}\n".format(best_checkpoint))
# tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
# model = model_class.from_pretrained(best_checkpoint)
# model.to(args.device)
output_test_results_file = os.path.join(args.output_dir,
"test_results.txt")
with open(output_test_results_file, "a") as result_writer:
for lang in args.predict_langs.split(','):
if not os.path.exists(
os.path.join(args.data_dir, lang, 'test.{}'.format(
args.model_name_or_path))):
logger.info("Language {} does not exist".format(lang))
continue
result, predictions = evaluate(args,
model,
tokenizer,
labels,
pad_token_label_id,
mode="test",
lang=lang,
lang2id=lang2id)
# Save results
result_writer.write(
"=====================\nlanguage={}\n".format(lang))
for key in sorted(result.keys()):
result_writer.write("{} = {}\n".format(
key, str(result[key])))
# Save predictions
output_test_predictions_file = os.path.join(
args.output_dir, "test_{}_predictions.txt".format(lang))
infile = os.path.join(
args.data_dir, lang,
"test.{}".format(args.model_name_or_path))
idxfile = infile + '.idx'
save_predictions(args, predictions,
output_test_predictions_file, infile, idxfile)
# Predict dev set
if args.do_predict_dev and args.local_rank in [-1, 0]:
# logger.info("Loading the best checkpoint from {}\n".format(best_checkpoint))
# tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
# model = model_class.from_pretrained(best_checkpoint)
# model.to(args.device)
output_test_results_file = os.path.join(args.output_dir,
"dev_results.txt")
with open(output_test_results_file, "w") as result_writer:
for lang in args.predict_langs.split(','):
if not os.path.exists(
os.path.join(args.data_dir, lang, 'dev.{}'.format(
args.model_name_or_path))):
logger.info("Language {} does not exist".format(lang))
continue
result, predictions = evaluate(args,
model,
tokenizer,
labels,
pad_token_label_id,
mode="dev",
lang=lang,
lang2id=lang2id)
# Save results
result_writer.write(
"=====================\nlanguage={}\n".format(lang))
for key in sorted(result.keys()):
result_writer.write("{} = {}\n".format(
key, str(result[key])))
# Save predictions
output_test_predictions_file = os.path.join(
args.output_dir, "dev_{}_predictions.txt".format(lang))
infile = os.path.join(args.data_dir, lang,
"dev.{}".format(args.model_name_or_path))
idxfile = infile + '.idx'
save_predictions(args, predictions,
output_test_predictions_file, infile, idxfile)
def main():
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument(
"--data_dir",
default=None,
type=str,
required=True,
help=
"The input data dir. Should contain the training files for the NER/POS task."
)
parser.add_argument("--dataset_name",
required=True,
type=str,
help="Name of the dataset")
parser.add_argument("--model_type",
default=None,
type=str,
required=True,
help="Model type selected in the list: " +
", ".join(MODEL_CLASSES.keys()))
parser.add_argument("--model_name_or_path",
default=None,
type=str,
required=True,
help="Path to pre-trained model or shortcut name.")
parser.add_argument(
"--output_dir",
default=None,
type=str,
required=True,
help=
"The output directory where the model predictions and checkpoints will be written."
)
## Other parameters
parser.add_argument(
"--labels",
default="",
type=str,
help=
"Path to a file containing all labels. If not specified, NER/POS labels are used."
)
parser.add_argument(
"--config_name",
default="",
type=str,
help="Pretrained config name or path if not the same as model_name")
parser.add_argument(
"--tokenizer_name",
default="",
type=str,
help="Pretrained tokenizer name or path if not the same as model_name")
parser.add_argument(
"--cache_dir",
default=None,
type=str,
help=
"Where do you want to store the pre-trained models downloaded from s3")
parser.add_argument(
"--max_seq_length",
default=128,
type=int,
help=
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded.")
parser.add_argument("--do_train",
action="store_true",
help="Whether to run training.")
parser.add_argument("--do_eval",
action="store_true",
help="Whether to run eval on the dev set.")
parser.add_argument("--do_predict",
action="store_true",
help="Whether to run predictions on the test set.")
parser.add_argument("--do_predict_dev",
action="store_true",
help="Whether to run predictions on the dev set.")
parser.add_argument("--init_checkpoint",
default=None,
type=str,
help="initial checkpoint for train/predict")
parser.add_argument(
"--evaluate_during_training",
action="store_true",
help="Whether to run evaluation during training at each logging step.")
parser.add_argument(
"--do_lower_case",
action="store_true",
help="Set this flag if you are using an uncased model.")
parser.add_argument("--few_shot",
default=-1,
type=int,
help="num of few-shot exampes")
parser.add_argument("--per_gpu_train_batch_size",
default=8,
type=int,
help="Batch size per GPU/CPU for training.")
parser.add_argument("--per_gpu_eval_batch_size",
default=8,
type=int,
help="Batch size per GPU/CPU for evaluation.")
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help=
"Number of updates steps to accumulate before performing a backward/update pass."
)
parser.add_argument("--learning_rate",
default=5e-5,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--weight_decay",
default=0.0,
type=float,
help="Weight decay if we apply some.")
parser.add_argument("--adam_epsilon",
default=1e-8,
type=float,
help="Epsilon for Adam optimizer.")
parser.add_argument("--max_grad_norm",
default=1.0,
type=float,
help="Max gradient norm.")
parser.add_argument("--num_train_epochs",
default=3.0,
type=float,
help="Total number of training epochs to perform.")
parser.add_argument(
"--max_steps",
default=-1,
type=int,
help=
"If > 0: set total number of training steps to perform. Override num_train_epochs."
)
parser.add_argument("--warmup_steps",
default=0,
type=int,
help="Linear warmup over warmup_steps.")
parser.add_argument("--logging_steps",
type=int,
default=50,
help="Log every X updates steps.")
parser.add_argument("--save_steps",
type=int,
default=50,
help="Save checkpoint every X updates steps.")
parser.add_argument("--save_only_best_checkpoint",
action="store_true",
help="Save only the best checkpoint during training")
parser.add_argument(
"--eval_all_checkpoints",
action="store_true",
help=
"Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number"
)
parser.add_argument("--no_cuda",
action="store_true",
help="Avoid using CUDA when available")
parser.add_argument("--overwrite_output_dir",
action="store_true",
help="Overwrite the content of the output directory")
parser.add_argument(
"--overwrite_cache",
action="store_true",
help="Overwrite the cached training and evaluation sets")
parser.add_argument("--seed",
type=int,
default=42,
help="random seed for initialization")
parser.add_argument("--workers",
type=int,
default=20,
help="number of workers for data loading.")
parser.add_argument(
"--fp16",
action="store_true",
help=
"Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit"
)
parser.add_argument(
"--fp16_opt_level",
type=str,
default="O1",
help=
"For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
"See details at https://nvidia.github.io/apex/amp.html")
parser.add_argument("--local_rank",
type=int,
default=-1,
help="For distributed training: local_rank")
parser.add_argument("--server_ip",
type=str,
default="",
help="For distant debugging.")
parser.add_argument("--server_port",
type=str,
default="",
help="For distant debugging.")
parser.add_argument("--log_file", type=str, default=None, help="log file")
parser.add_argument(
"--eval_patience",
type=int,
default=-1,
help=
"wait N times of decreasing dev score before early stop during training"
)
args = parser.parse_args()
if os.path.exists(args.output_dir) and os.listdir(
args.output_dir
) and args.do_train and not args.overwrite_output_dir:
raise ValueError(
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome."
.format(args.output_dir))
# Setup distant debugging if needed
if args.server_ip and args.server_port:
import ptvsd
print("Waiting for debugger attach")
ptvsd.enable_attach(address=(args.server_ip, args.server_port),
redirect_output=True)
ptvsd.wait_for_attach()
# Setup CUDA, GPU & distributed training
if args.local_rank == -1 or args.no_cuda:
device = torch.device("cuda" if torch.cuda.is_available()
and not args.no_cuda else "cpu")
args.n_gpu = torch.cuda.device_count()
else:
# Initializes the distributed backend which sychronizes nodes/GPUs
torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank)
torch.distributed.init_process_group(backend="nccl")
args.n_gpu = 1
args.device = device
# Setup logging
logging.basicConfig(
handlers=[logging.FileHandler(args.log_file),
logging.StreamHandler()],
format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S',
level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN)
logging.info("Input args: %r" % args)
logger.warning(
"Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
args.local_rank, device, args.n_gpu, bool(args.local_rank != -1),
args.fp16)
# Set seed
set_seed(args)
# Prepare NER/POS task
labels = get_labels()
num_labels = len(labels)
# Use cross entropy ignore index as padding label id
# so that only real label ids contribute to the loss later
pad_token_label_id = CrossEntropyLoss().ignore_index
# Load pretrained model and tokenizer
# Make sure only the first process in distributed training loads model/vocab
if args.local_rank not in [-1, 0]:
torch.distributed.barrier()
args.model_type = args.model_type.lower()
config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]
config = config_class.from_pretrained(
args.config_name if args.config_name else args.model_name_or_path,
num_labels=num_labels,
cache_dir=args.cache_dir if args.cache_dir else None)
tokenizer = tokenizer_class.from_pretrained(
args.tokenizer_name
if args.tokenizer_name else args.model_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=args.cache_dir if args.cache_dir else None)
if args.init_checkpoint:
logger.info("loading from init_checkpoint={}".format(
args.init_checkpoint))
model = model_class.from_pretrained(args.init_checkpoint,
config=config,
cache_dir=args.init_checkpoint)
else:
logger.info("loading from cached model = {}".format(
args.model_name_or_path))
model = model_class.from_pretrained(
args.model_name_or_path,
from_tf=bool(".ckpt" in args.model_name_or_path),
config=config,
cache_dir=args.cache_dir if args.cache_dir else None)
# Make sure only the first process in distributed training loads model/vocab
if args.local_rank == 0:
torch.distributed.barrier()
model.to(args.device)
logger.info("Training/evaluation parameters %s", args)
# Training
if args.do_train:
train_dataset = load_and_cache_examples(args,
tokenizer,
labels,
pad_token_label_id,
mode="train",
few_shot=args.few_shot)
global_step, tr_loss = train(args, train_dataset, model, tokenizer,
labels, pad_token_label_id)
logger.info(" global_step = %s, average loss = %s", global_step,
tr_loss)
# Saving best-practices: if you use default names for the model,
# you can reload it using from_pretrained()
if args.do_train and (args.local_rank == -1
or torch.distributed.get_rank() == 0):
# Create output directory if needed
if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
os.makedirs(args.output_dir)
# Save model, configuration and tokenizer using `save_pretrained()`.
# They can then be reloaded using `from_pretrained()`
# Take care of distributed/parallel training
logger.info("Saving model checkpoint to %s", args.output_dir)
model_to_save = model.module if hasattr(model, "module") else model
model_to_save.save_pretrained(args.output_dir)
tokenizer.save_pretrained(args.output_dir)
# Good practice: save your training arguments together with the model
torch.save(args, os.path.join(args.output_dir, "training_args.bin"))
# Initialization for evaluation
results = {}
if args.init_checkpoint:
best_checkpoint = args.init_checkpoint
elif os.path.exists(os.path.join(args.output_dir, 'checkpoint-best')):
best_checkpoint = os.path.join(args.output_dir, 'checkpoint-best')
else:
best_checkpoint = args.output_dir
best_f1 = 0
# Evaluation
if args.do_eval and args.local_rank in [-1, 0]:
tokenizer = tokenizer_class.from_pretrained(
args.output_dir, do_lower_case=args.do_lower_case)
checkpoints = [args.output_dir]
if args.eval_all_checkpoints:
checkpoints = list(
os.path.dirname(c) for c in sorted(
glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME,
recursive=True)))
logging.getLogger("pytorch_transformers.modeling_utils").setLevel(
logging.WARN)
logger.info("Evaluate the following checkpoints: %s", checkpoints)
for checkpoint in checkpoints:
global_step = checkpoint.split(
"-")[-1] if len(checkpoints) > 1 else ""
model = model_class.from_pretrained(checkpoint)
model.to(args.device)
result, _ = evaluate(args,
model,
tokenizer,
labels,
pad_token_label_id,
mode="valid",
prefix=global_step)
if result["f1"] > best_f1:
best_checkpoint = checkpoint
best_f1 = result["f1"]
if global_step:
result = {
"{}_{}".format(global_step, k): v
for k, v in result.items()
}
results.update(result)
output_eval_file = os.path.join(args.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
for key in sorted(results.keys()):
writer.write("{} = {}\n".format(key, str(results[key])))
writer.write("best checkpoint = {}, best f1 = {}\n".format(
best_checkpoint, best_f1))
# Prediction
if args.do_predict and args.local_rank in [-1, 0]:
logger.info(
"Loading the best checkpoint from {}\n".format(best_checkpoint))
tokenizer = tokenizer_class.from_pretrained(
args.output_dir, do_lower_case=args.do_lower_case)
model = model_class.from_pretrained(best_checkpoint)
model.to(args.device)
if args.dataset_name:
output_test_results_file = os.path.join(
args.output_dir, "{}_results.txt".format(args.dataset_name))
output_test_predictions_file = os.path.join(
args.output_dir,
"{}_predictions.txt".format(args.dataset_name))
else:
output_test_results_file = os.path.join(args.output_dir,
"test_results.txt")
output_test_predictions_file = os.path.join(
args.output_dir, "test_predictions.txt")
with open(output_test_results_file, "w") as result_writer:
result, predictions = evaluate(args,
model,
tokenizer,
labels,
pad_token_label_id,
mode="test",
print_result=False)
# Save results
logger.info("***** Test results *****")
for key, value in result.items():
logger.info(" %s = %s", key, value)
result_writer.write("%s = %s\n" % (key, value))
# Save predictions
with open(output_test_predictions_file, "w") as writer:
with open(os.path.join(args.data_dir, "test.txt"), "r") as f:
example_id = 0
for line in f:
ex = json.loads(line.strip())
# note. we may truncate source
assert len(ex['source']) >= len(
predictions[example_id])
keyphrases = []
kp_tokens = []
for idx, tag in enumerate(predictions[example_id]):
if tag == 'B':
if len(kp_tokens) != 0:
# save the previous keyphrase
keyphrases.append(' '.join(kp_tokens))
kp_tokens = []
kp_tokens.append(ex['source'][idx])
elif tag == 'I':
# note. model may predict I tag which is not preceded by B tag
# assert len(kp_tokens) > 0
kp_tokens.append(ex['source'][idx])
else:
if len(kp_tokens) != 0:
keyphrases.append(' '.join(kp_tokens))
kp_tokens = []
if len(kp_tokens) != 0:
keyphrases.append(' '.join(kp_tokens))
# removing duplicates
kps = []
[kps.append(kp) for kp in keyphrases if kp not in kps]
writer.write(';'.join(kps) + '\n')
example_id += 1
# Predict dev set
if args.do_predict_dev and args.local_rank in [-1, 0]:
logger.info(
"Loading the best checkpoint from {}\n".format(best_checkpoint))
tokenizer = tokenizer_class.from_pretrained(
args.output_dir, do_lower_case=args.do_lower_case)
model = model_class.from_pretrained(best_checkpoint)
model.to(args.device)
output_test_results_file = os.path.join(args.output_dir,
"eval_results.txt")
with open(output_test_results_file, "w") as result_writer:
result, predictions = evaluate(args,
model,
tokenizer,
labels,
pad_token_label_id,
mode="valid")
# Save results
logger.info("***** Eval results *****")
for key, value in result.items():
logger.info(" %s = %s", key, value)
result_writer.write("%s = %s\n" % (key, value))