def run(model_type, task, ori_task=None, gpu="-1"):
gpu = str(gpu)
task = task.lower()
ori_task = ori_task.lower() if ori_task is not None else None
os.environ["CUDA_VISIBLE_DEVICES"] = gpu
if model_type in ["Glove", "Word2vec", "LSTM"]:
print("{} not supported now.".format(model_type))
else:
_model_type = model_type.split('-')[0]
ori_model_name = "{}[{}]".format(
model_type, ori_task) if ori_task is not None else None
_model_name_or_path = model_type if ori_task is None else "./model/{}".format(
ori_model_name)
_task_name = find_source_task(task)
_data_dir = "./data/{}".format(task.upper())
_max_seq_length = str(128)
_per_gpu_train_batch_size = str(8)
_learning_rate = "2e-5" if ori_task is None else "0.0"
_num_train_epochs = str(3)
new_model_name = "{}[{}]".format(
model_type, task) if ori_task is None else "{}[{}]-[{}]".format(
model_type, task, ori_task)
_output_dir = "./model/{}".format(new_model_name)
main([
"--model_type", _model_type, "--model_name_or_path",
_model_name_or_path, "--task_name", _task_name, "--do_train",
"--do_eval", "--do_lower_case", "--data_dir", _data_dir,
"--max_seq_length", _max_seq_length, "--per_gpu_train_batch_size",
_per_gpu_train_batch_size, "--learning_rate", _learning_rate,
"--num_train_epochs", _num_train_epochs, "--output_dir",
_output_dir, "--save_steps", "10000", "--overwrite_output_dir"
])
return new_model_name
def test_run_glue(self):
stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)
tmp_dir = self.get_auto_remove_tmp_dir()
testargs = f"""
run_glue.py
--model_name_or_path distilbert-base-uncased
--output_dir {tmp_dir}
--overwrite_output_dir
--train_file ./tests/fixtures/tests_samples/MRPC/train.csv
--validation_file ./tests/fixtures/tests_samples/MRPC/dev.csv
--do_train
--do_eval
--per_device_train_batch_size=2
--per_device_eval_batch_size=1
--learning_rate=1e-4
--max_steps=10
--warmup_steps=2
--seed=42
--max_seq_length=128
""".split()
if is_cuda_and_apex_available():
testargs.append("--fp16")
with patch.object(sys, "argv", testargs):
run_glue.main()
result = get_results(tmp_dir)
self.assertGreaterEqual(result["eval_accuracy"], 0.75)
def test_run_glue(self):
stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)
testargs = """
run_glue.py
--model_name_or_path bert-base-uncased
--data_dir ./tests/fixtures/tests_samples/MRPC/
--task_name mrpc
--do_train
--do_eval
--output_dir ./tests/fixtures/tests_samples/temp_dir
--per_gpu_train_batch_size=2
--per_gpu_eval_batch_size=1
--learning_rate=1e-4
--max_steps=10
--warmup_steps=2
--overwrite_output_dir
--seed=42
--max_seq_length=128
""".split()
with patch.object(sys, "argv", testargs):
result = run_glue.main()
del result["loss"]
for value in result.values():
self.assertGreaterEqual(value, 0.75)
def test_run_glue(self):
stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)
testargs = [
"run_glue.py",
"--data_dir=./examples/tests_samples/MRPC/",
"--task_name=mrpc",
"--do_train",
"--do_eval",
"--output_dir=./examples/tests_samples/temp_dir",
"--per_gpu_train_batch_size=2",
"--per_gpu_eval_batch_size=1",
"--learning_rate=1e-4",
"--max_steps=10",
"--warmup_steps=2",
"--overwrite_output_dir",
"--seed=42",
]
model_type, model_name = ("--model_type=bert",
"--model_name_or_path=bert-base-uncased")
with patch.object(sys, "argv", testargs + [model_type, model_name]):
result = run_glue.main()
for value in result.values():
self.assertGreaterEqual(value, 0.75)
def test_run_glue(self):
stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)
tmp_dir = self.get_auto_remove_tmp_dir()
testargs = f"""
run_glue.py
--model_name_or_path distilbert-base-uncased
--data_dir ./tests/fixtures/tests_samples/MRPC/
--output_dir {tmp_dir}
--overwrite_output_dir
--task_name mrpc
--do_train
--do_eval
--per_device_train_batch_size=2
--per_device_eval_batch_size=1
--learning_rate=1e-4
--max_steps=10
--warmup_steps=2
--seed=42
--max_seq_length=128
"""
output_dir = "./tests/fixtures/tests_samples/temp_dir_{}".format(
hash(testargs))
testargs += "--output_dir " + output_dir
testargs = testargs.split()
if is_cuda_and_apex_available():
testargs.append("--fp16")
with patch.object(sys, "argv", testargs):
result = run_glue.main()
del result["eval_loss"]
for value in result.values():
self.assertGreaterEqual(value, 0.75)
def execute(self):
flags = copy.deepcopy(_args)
cur_process = multiprocessing.current_process()
flags.output_dir = os.path.join(flags.output_dir)
#flags.output_dir = os.path.join(flags.output_dir, cur_process.name)
flags.job_id = self._job_id
print("flags.output_dir: ", flags.output_dir)
self.update_flags(self._settings, flags)
os.environ["CUDA_VISIBLE_DEVICES"] = self._gpu_id
# os.environ["TF_CPP_MIN_LOG_LEVEL"] = '3'
self.print_user_flags(flags)
self.result, self.all_result = main(flags)
parser.add_argument('-output_path', type=str, required=True)
parser.add_argument('-alignment_model_path', type=str, required=True)
parser.add_argument('-database', type=str, default='None')
args = parser.parse_args()
aligner = docSum2MRPC_Aligner(data_path=args.data_path, mode=args.mode,
log_file=args.log_file, output_file = args.output_path,
database=args.database)
logging.info(f'output_file_name: {args.output_file}')
summary_files = glob.glob(f"{args.data_path}/summaries/*")
for sfile in summary_files:
print ('Starting with summary {}'.format(sfile))
aligner.read_and_split(args.database, sfile)
aligner.scu_span_aligner()
aligner.save_predictions()
with redirect_argv('python --model_type roberta --model_name_or_path roberta-large-mnli --task_name MRPC --do_eval'
f' --calc_final_alignments --weight_decay 0.1 --data_dir {args.output_path}'
' --max_seq_length 128 --per_gpu_train_batch_size 16 --per_gpu_eval_batch_size 16 --learning_rate 2e-6'
' --logging_steps 500 --num_train_epochs 2.0 --evaluate_during_training --overwrite_cache'
f' --output_dir {args.alignment_model_path}'):
run_glue.main()