answerdotai/answerai-colbert-small-v1
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
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 3.9K
Corpus p50 203.2ms
Query TPS 186
Query p50 300.7ms
Performance L4 b1 c16
Corpus TPS 37.3K
Corpus p50 54.7ms
Query TPS 3.6K
Query p50 47.1ms
default
Quality
ndcg at 10 0.4154
map at 10 0.3645
mrr at 10 0.4213
Performance L4 b1 c16
Corpus TPS 44.9K
Corpus p50 45.2ms
Query TPS 4.5K
Query p50 37.7ms
CosQA
Code search with natural language queries
Corpus: 6,267 Queries: 500
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 1.1K
Corpus p50 345.4ms
Query TPS 102
Query p50 466.2ms
Performance L4 b1 c16
Corpus TPS 15.7K
Corpus p50 53.9ms
Query TPS 2.0K
Query p50 47.4ms
default
Quality
ndcg at 10 0.2844
map at 10 0.2180
mrr at 10 0.2069
Performance L4 b1 c16
Corpus TPS 19.0K
Corpus p50 43.5ms
Query TPS 2.3K
Query p50 40.6ms
FiQA2018
Financial opinion mining and question answering
Corpus: 57,599 Queries: 648
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 3.4K
Corpus p50 384.8ms
Query TPS 174
Query p50 547.3ms
Performance L4 b1 c16
Corpus TPS 43.1K
Corpus p50 59.1ms
Query TPS 3.7K
Query p50 50.0ms
default
Quality
ndcg at 10 0.4103
map at 10 0.3338
mrr at 10 0.4965
Performance L4 b1 c16
Corpus TPS 49.3K
Corpus p50 47.9ms
Query TPS 4.5K
Query p50 40.0ms
LegalBenchConsumerContractsQA
Question answering on consumer contracts
Corpus: 153 Queries: 396
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 11.2K
Corpus p50 286.1ms
Query TPS 254
Query p50 500.2ms
Performance L4 b1 c16
Corpus TPS 83.1K
Corpus p50 83.9ms
Query TPS 4.8K
Query p50 52.2ms
default
Quality
ndcg at 10 0.7840
map at 10 0.7315
mrr at 10 0.7315
Performance L4 b1 c16
Corpus TPS 115.6K
Corpus p50 62.7ms
Query TPS 6.4K
Query p50 40.8ms
NFCorpus
Biomedical literature search from NutritionFacts.org
Corpus: 3,593 Queries: 323
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 5.7K
Corpus p50 300.1ms
Query TPS 210
Query p50 178.7ms
Performance L4 b1 c16
Corpus TPS 60.7K
Corpus p50 69.4ms
Query TPS 1.4K
Query p50 52.3ms
default
Quality
ndcg at 10 0.3715
map at 10 0.1440
mrr at 10 0.5870
Performance L4 b1 c16
Corpus TPS 75.8K
Corpus p50 55.4ms
Query TPS 1.9K
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 TPS 43.6K
Corpus p50 44.3ms
Query TPS 4.0K
Query p50 35.0ms
SCIDOCS
Citation prediction, document classification, and recommendation for scientific papers
Corpus: 25,656 Queries: 1,000
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 2.9K
Corpus p50 501.5ms
Query TPS 222
Query p50 353.0ms
Performance L4 b1 c16
Corpus TPS 48.4K
Corpus p50 57.9ms
Query TPS 3.8K
Query p50 46.7ms
default
Quality
ndcg at 10 0.1778
map at 10 0.1046
mrr at 10 0.3078
Performance L4 b1 c16
Corpus TPS 59.1K
Corpus p50 47.2ms
Query TPS 4.7K
Query p50 38.2ms
SciFact
Scientific claim verification using research literature
Corpus: 5,183 Queries: 300
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 4.9K
Corpus p50 443.1ms
Query TPS 332
Query p50 430.3ms
Performance L4 b1 c16
Corpus TPS 61.5K
Corpus p50 63.3ms
Query TPS 5.2K
Query p50 50.5ms
default
Quality
ndcg at 10 0.7405
map at 10 0.7015
mrr at 10 0.7120
Performance L4 b1 c16
Corpus TPS 75.3K
Corpus p50 51.0ms
Query TPS 6.7K
Query p50 39.6ms
StackOverflowQA
Programming question answering from Stack Overflow
Corpus: 19,931 Queries: 1,994
default_candidates-k-50_candidates-model-Alibaba-NLP__gte-multilingual-base
Performance L4-SPOT b1 c16
Corpus TPS 4.4K
Corpus p50 379.7ms
Query TPS 5.2K
Query p50 432.3ms
Performance L4 b1 c16
Corpus TPS 55.3K
Corpus p50 60.2ms
Query TPS 73.2K
Query p50 64.7ms
default
Quality
ndcg at 10 0.5461
map at 10 0.5130
mrr at 10 0.5130
Performance L4 b1 c16
Corpus TPS 62.1K
Corpus p50 53.0ms
Query TPS 88.2K
Query p50 52.4ms