help=f"Format of output. Available: {[x.value for x in list(OutputFormat)]}")
run_parser.add_argument('--output', type=str, metavar='path', required=False, help="Path to output file.")
run_parser.add_argument('--max-passage', action='store_true',
default=False, help="Select only max passage from document.")
run_parser.add_argument('--max-passage-hits', type=int, metavar='num', required=False, default=100,
help="Final number of hits when selecting only max passage.")
run_parser.add_argument('--max-passage-delimiter', type=str, metavar='str', required=False, default='#',
help="Delimiter between docid and passage id.")
run_parser.add_argument('--batch-size', type=int, metavar='num', required=False,
default=1, help="Specify batch size to search the collection concurrently.")
run_parser.add_argument('--threads', type=int, metavar='num', required=False,
default=1, help="Maximum number of threads to use.")
args = parse_args(parser, commands)
query_iterator = get_query_iterator(args.run.topics, TopicsFormat(args.run.topics_format))
topics = query_iterator.topics
query_encoder = init_query_encoder(args.dense.encoder,
args.dense.tokenizer,
args.run.topics,
args.dense.encoded_queries,
args.dense.device)
if os.path.exists(args.dense.index):
# create searcher from index directory
dsearcher = SimpleDenseSearcher(args.dense.index, query_encoder)
else:
# create searcher from prebuilt index name
dsearcher = SimpleDenseSearcher.from_prebuilt_index(args.dense.index, query_encoder)
required=False,
default=1,
help="Specify batch size to search the collection concurrently.")
parser.add_argument('--threads',
type=int,
metavar='num',
required=False,
default=1,
help="Maximum number of threads to use.")
parser.add_argument('--tokenizer',
type=str,
help='tokenizer used to preprocess topics')
args = parser.parse_args()
query_iterator = get_query_iterator(args.topics,
TopicsFormat(args.topics_format))
topics = query_iterator.topics
if os.path.exists(args.index):
# create searcher from index directory
searcher = SimpleSearcher(args.index)
else:
# create searcher from prebuilt index name
searcher = SimpleSearcher.from_prebuilt_index(args.index)
if args.language != 'en':
searcher.set_language(args.language)
if not searcher:
exit()
parser.add_argument('--encoder',
type=str,
help='encoder name or path',
default='facebook/dpr-question_encoder-multiset-base',
required=False)
parser.add_argument('--output',
type=str,
help='path to store query embeddings',
required=True)
parser.add_argument('--device',
type=str,
help='device cpu or cuda [cuda:0, cuda:1...]',
default='cpu',
required=False)
args = parser.parse_args()
query_iterator = get_query_iterator(
args.topics, TopicsFormat(TopicsFormat.DEFAULT.value))
topics = query_iterator.topics
encoder = DkrrDprQueryEncoder(args.encoder, args.device)
embeddings = {'id': [], 'text': [], 'embedding': []}
for index, (topic_id, text) in enumerate(
tqdm(query_iterator, total=len(topics.keys()))):
embeddings['id'].append(topic_id)
embeddings['text'].append(text)
embeddings['embedding'].append(encoder.encode(text))
embeddings = pd.DataFrame(embeddings)
embeddings.to_pickle(args.output)