if "eval" in args.mode:
if not os.path.exists(args.eval_file):
raise FileNotFoundError(
"eval data not found. Datasets can be obtained using examples/nlp/question_answering/get_squad.py"
)
if "test" in args.mode:
if not os.path.exists(args.test_file):
raise FileNotFoundError(
"test data not found. Datasets can be obtained using examples/nlp/question_answering/get_squad.py"
)
# Instantiate neural factory with supported backend
nf = nemo_core.NeuralModuleFactory(
local_rank=args.local_rank,
optimization_level=args.amp_opt_level,
log_dir=args.work_dir,
create_tb_writer=True,
files_to_copy=[__file__],
add_time_to_log_dir=False,
)
model = nemo_nlp.nm.trainables.get_pretrained_lm_model(
pretrained_model_name=args.pretrained_model_name,
config=args.bert_config,
vocab=args.vocab_file,
checkpoint=args.bert_checkpoint,
)
tokenizer = nemo.collections.nlp.data.tokenizers.get_tokenizer(
tokenizer_name=args.tokenizer,
pretrained_model_name=args.pretrained_model_name,
tokenizer_model=args.tokenizer_model,
if not exists(abs_data_dir):
raise ValueError(f"Folder `{abs_data_dir}` not found")
# Prepare the experiment (output) dir.
abs_work_dir = f'{expanduser(args.work_dir)}/dst_trade/'
logging.info(
"Logging the results of the experiment to: `{}`".format(abs_work_dir))
print("abs_data_dir = ", abs_data_dir)
data_desc = MultiWOZDataDesc(abs_data_dir, domains)
nf = nemo_core.NeuralModuleFactory(
backend=nemo_core.Backend.PyTorch,
local_rank=args.local_rank,
optimization_level=args.amp_opt_level,
log_dir=abs_work_dir,
create_tb_writer=True,
files_to_copy=[__file__],
add_time_to_log_dir=True,
)
vocab_size = len(data_desc.vocab)
encoder = EncoderRNN(vocab_size, args.emb_dim, args.hid_dim, args.dropout,
args.n_layers)
decoder = TRADEGenerator(
data_desc.vocab,
encoder.embedding,
args.hid_dim,
args.dropout,
data_desc.slots,
