if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Compute embeddings for KILT topics')
parser.add_argument('--topics', required=True)
parser.add_argument('--output', default="embedding.pkl", help="Name and path to output file.")
parser.add_argument('--encoder', metavar='path to query encoder checkpoint or encoder name',
required=True,
help="Path to query encoder pytorch checkpoint or hgf encoder model name")
parser.add_argument('--tokenizer', metavar='name or path',
required=True,
help="Path to a hgf tokenizer name or path")
parser.add_argument('--device', metavar='device to run query encoder', required=False, default='cpu',
help="Device to run query encoder, cpu or [cuda:0, cuda:1, ...]")
args = parser.parse_args()
query_iterator = get_query_iterator(args.topics, TopicsFormat.KILT)
query_encoder = DprQueryEncoder(encoder_dir=args.encoder, tokenizer_name=args.tokenizer, device=args.device)
texts = []
embeddings = []
for i, (topic_id, text) in enumerate(tqdm(query_iterator)):
texts.append(text)
embeddings.append(query_encoder.encode(text))
df = pd.DataFrame({
'text': texts,
'embedding': embeddings
})
df.to_pickle(args.output)
metavar='num',
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)
