answerdotai/answerai-colbert-small-v1 (Encode)
answerai-colbert-small-v1 is a new, proof-of-concept model by Answer.AI, showing the strong performance multi-vector models with the new JaColBERTv2.5 training recipe and some extra tweaks can reach, even with just 33 million parameters.
Benchmarks
CQADupstackPhysicsRetrieval
Duplicate question retrieval from StackExchange Physics
Corpus: 38,314 Queries: 1,039
Quality
ndcg at 10 0.4154
map at 10 0.3645
mrr at 10 0.4213
Performance L4 b1 c16
Corpus 44.9K tok/s
Corpus p50 45.2ms
Query 4.5K tok/s
Query p50 37.7ms
CosQA
Code search with natural language queries
Corpus: 6,267 Queries: 500
Quality
ndcg at 10 0.2844
map at 10 0.2180
mrr at 10 0.2069
Performance L4 b1 c16
Corpus 19.0K tok/s
Corpus p50 43.5ms
Query 2.3K tok/s
Query p50 40.6ms
FiQA2018
Financial opinion mining and question answering
Corpus: 57,599 Queries: 648
Quality
ndcg at 10 0.4103
map at 10 0.3338
mrr at 10 0.4965
Performance L4 b1 c16
Corpus 49.3K tok/s
Corpus p50 47.9ms
Query 4.5K tok/s
Query p50 40.0ms
LegalBenchConsumerContractsQA
Question answering on consumer contracts
Corpus: 153 Queries: 396
Quality
ndcg at 10 0.7840
map at 10 0.7315
mrr at 10 0.7315
Performance L4 b1 c16
Corpus 115.6K tok/s
Corpus p50 62.7ms
Query 6.4K tok/s
Query p50 40.8ms
NFCorpus
Biomedical literature search from NutritionFacts.org
Corpus: 3,593 Queries: 323
Quality
ndcg at 10 0.3715
map at 10 0.1440
mrr at 10 0.5870
Performance L4 b1 c16
Corpus 75.8K tok/s
Corpus p50 55.4ms
Query 1.9K tok/s
Query p50 41.4ms
NanoFiQA2018Retrieval
Smaller subset of the FiQA financial QA dataset
Quality
ndcg at 10 0.5563
map at 10 0.4718
mrr at 10 0.6192
Performance L4 b1 c16
Corpus 43.6K tok/s
Corpus p50 44.3ms
Query 4.0K tok/s
Query p50 35.0ms
SCIDOCS
Citation prediction, document classification, and recommendation for scientific papers
Corpus: 25,656 Queries: 1,000
Quality
ndcg at 10 0.1778
map at 10 0.1046
mrr at 10 0.3078
Performance L4 b1 c16
Corpus 59.1K tok/s
Corpus p50 47.2ms
Query 4.7K tok/s
Query p50 38.2ms
SciFact
Scientific claim verification using research literature
Corpus: 5,183 Queries: 300
Quality
ndcg at 10 0.7405
map at 10 0.7015
mrr at 10 0.7120
Performance L4 b1 c16
Corpus 75.3K tok/s
Corpus p50 51.0ms
Query 6.7K tok/s
Query p50 39.6ms
StackOverflowQA
Programming question answering from Stack Overflow
Corpus: 19,931 Queries: 1,994
Quality
ndcg at 10 0.5461
map at 10 0.5130
mrr at 10 0.5130
Performance L4 b1 c16
Corpus 62.1K tok/s
Corpus p50 53.0ms
Query 88.2K tok/s
Query p50 52.4ms