def make_binary_da_dataset(input_prefix, output_prefix, lang, num_workers,
da_mapping):
logger.info("Adding domain indexes")
n_seq_tok = [0, 0]
replaced = Counter()
def merge_result(worker_result):
n_seq_tok[0] += worker_result["nseq"]
input_file = "{}{}".format(input_prefix,
("." + lang) if lang is not None else "")
offsets = Binarizer.find_offsets(input_file, num_workers)
pool = None
# TODO: Error encounters if num_workers>1:
# No such file or directory: 'data-bin/iwslt14.tokenized.de-en/train.da1.en-de.en.idx'
if num_workers > 1:
pool = Pool(processes=num_workers - 1)
for worker_id in range(1, num_workers):
prefix = "{}{}".format(output_prefix, worker_id)
pool.apply_async(
binarize_da,
(
args,
input_file,
prefix,
lang,
da_mapping,
offsets[worker_id],
offsets[worker_id + 1],
),
callback=merge_result,
)
pool.close()
ds = indexed_dataset.make_builder(
dataset_dest_file(args, output_prefix, lang, "bin"),
impl=args.dataset_impl,
)
merge_result(
Binarizer.binarize_da(input_file,
lambda t: ds.add_item(t),
da_mapping,
offset=0,
end=offsets[1]))
if num_workers > 1:
pool.join()
for worker_id in range(1, num_workers):
prefix = "{}{}".format(output_prefix, worker_id)
temp_file_path = dataset_dest_prefix(args, prefix, lang)
ds.merge_file_(temp_file_path)
os.remove(indexed_dataset.data_file_path(temp_file_path))
os.remove(indexed_dataset.index_file_path(temp_file_path))
ds.finalize(dataset_dest_file(args, output_prefix, lang, "idx"))
def binarize_da(args, filename, output_prefix, da_mapping, offset, end):
ds = indexed_dataset.make_builder(
dataset_dest_file(args, output_prefix, None, "bin"),
impl=args.dataset_impl,
vocab_size=None,
)
def consumer(tensor):
ds.add_item(tensor)
res = Binarizer.binarize_da(filename,
consumer,
da_mapping,
offset=offset,
end=end)
ds.finalize(dataset_dest_file(args, output_prefix, None, "idx"))
return res
