BAAI/bge-m3 (Encode)
For more details please refer to our github repo: https://github.com/FlagOpen/FlagEmbedding
Overview
Benchmarks
CQADupstackPhysicsRetrieval
Duplicate question retrieval from StackExchange Physics
Corpus: 38,314 Queries: 1,039
dense
Performance L4 b1 c16
Corpus 27.3K tok/s
Corpus p50 74.8ms
Query 3.1K tok/s
Query p50 45.2ms
sparse
Performance L4 b1 c16
Corpus 27.8K tok/s
Corpus p50 70.8ms
Query 3.3K tok/s
Query p50 44.9ms
multivector
Performance L4 b1 c16
Corpus 28.0K tok/s
Corpus p50 70.7ms
Query 3.2K tok/s
Query p50 44.7ms
CosQA
Code search with natural language queries
Corpus: 6,267 Queries: 500
dense
Performance L4 b1 c16
Corpus 15.8K tok/s
Corpus p50 51.3ms
Query 1.7K tok/s
Query p50 47.9ms
sparse
Performance L4 b1 c16
Corpus 16.9K tok/s
Corpus p50 51.0ms
Query 1.9K tok/s
Query p50 43.7ms
multivector
Performance L4 b1 c16
Corpus 15.3K tok/s
Corpus p50 51.4ms
Query 1.8K tok/s
Query p50 45.8ms
FiQA2018
Financial opinion mining and question answering
Corpus: 57,599 Queries: 648
dense
Performance L4 b1 c16
Corpus 28.6K tok/s
Corpus p50 87.3ms
Query 3.1K tok/s
Query p50 48.4ms
sparse
Performance L4 b1 c16
Corpus 30.2K tok/s
Corpus p50 81.3ms
Query 3.3K tok/s
Query p50 46.2ms
multivector
Performance L4 b1 c16
Corpus 33.8K tok/s
Corpus p50 77.8ms
Query 3.6K tok/s
Query p50 44.8ms
LegalBenchConsumerContractsQA
Question answering on consumer contracts
Corpus: 153 Queries: 396
dense
Performance L4 b1 c16
Corpus 39.5K tok/s
Corpus p50 210.7ms
Query 4.2K tok/s
Query p50 51.7ms
sparse
Performance L4 b1 c16
Corpus 40.6K tok/s
Corpus p50 208.1ms
Query 5.1K tok/s
Query p50 45.0ms
multivector
Performance L4 b1 c16
Corpus 39.1K tok/s
Corpus p50 216.2ms
Query 5.1K tok/s
Query p50 44.0ms
NFCorpus
Biomedical literature search from NutritionFacts.org
Corpus: 3,593 Queries: 323
default
Quality
ndcg at 10 0.3144
map at 10 0.1174
mrr at 10 0.5243
Performance A10G b1 c4
Corpus 127 tok/s
Corpus p50 9.4s
Query 39 tok/s
Query p50 537.6ms
Performance L4 b1 c16
Corpus 38.1K tok/s
Corpus p50 100.0ms
Query 1.5K tok/s
Query p50 41.0ms
dense
Performance L4 b1 c16
Corpus 34.2K tok/s
Corpus p50 134.0ms
Query 937 tok/s
Query p50 64.2ms
sparse
Performance L4 b1 c16
Corpus 40.0K tok/s
Corpus p50 124.4ms
Query 1.4K tok/s
Query p50 45.0ms
multivector
Performance L4 b1 c16
Corpus 37.9K tok/s
Corpus p50 126.6ms
Query 1.5K tok/s
Query p50 46.1ms
NanoFiQA2018Retrieval
Smaller subset of the FiQA financial QA dataset
dense
Performance L4 b1 c16
Corpus 28.1K tok/s
Corpus p50 89.9ms
Query 2.5K tok/s
Query p50 59.2ms
default
Quality
ndcg at 10 0.5726
map at 10 0.4957
mrr at 10 0.6467
Performance L4 b1 c16
Corpus 31.2K tok/s
Corpus p50 68.9ms
Query 2.9K tok/s
Query p50 43.3ms
sparse
Performance L4 b1 c16
Corpus 30.3K tok/s
Corpus p50 87.5ms
Query 2.7K tok/s
Query p50 53.3ms
multivector
Performance L4 b1 c16
Corpus 30.6K tok/s
Corpus p50 86.6ms
Query 2.7K tok/s
Query p50 54.1ms
SCIDOCS
Citation prediction, document classification, and recommendation for scientific papers
Corpus: 25,656 Queries: 1,000
dense
Performance L4 b1 c16
Corpus 33.2K tok/s
Corpus p50 94.8ms
Query 2.5K tok/s
Query p50 54.2ms
sparse
Performance L4 b1 c16
Corpus 32.5K tok/s
Corpus p50 93.4ms
Query 3.4K tok/s
Query p50 44.7ms
multivector
Performance L4 b1 c16
Corpus 34.4K tok/s
Corpus p50 90.7ms
Query 3.4K tok/s
Query p50 43.0ms
bge_m3_flag
Quality
map at 10 0.0946
mrr at 10 0.2886
ndcg at 10 0.1636
default
Quality
map at 10 0.0946
mrr at 10 0.2886
ndcg at 10 0.1636
SciFact
Scientific claim verification using research literature
Corpus: 5,183 Queries: 300
dense
Performance L4 b1 c16
Corpus 37.9K tok/s
Corpus p50 118.2ms
Query 4.2K tok/s
Query p50 48.5ms
sparse
Performance L4 b1 c16
Corpus 37.8K tok/s
Corpus p50 117.6ms
Query 4.2K tok/s
Query p50 47.7ms
multivector
Performance L4 b1 c16
Corpus 38.2K tok/s
Corpus p50 116.5ms
Query 4.9K tok/s
Query p50 44.4ms
bge_m3_flag
Quality
map at 10 0.5942
mrr at 10 0.6102
ndcg at 10 0.6457
default
Quality
map at 10 0.5942
mrr at 10 0.6102
ndcg at 10 0.6457
StackOverflowQA
Programming question answering from Stack Overflow
Corpus: 19,931 Queries: 1,994
dense
Performance L4 b1 c16
Corpus 33.4K tok/s
Corpus p50 104.7ms
Query 33.8K tok/s
Query p50 138.5ms
sparse
Performance L4 b1 c16
Corpus 32.7K tok/s
Corpus p50 108.4ms
Query 34.8K tok/s
Query p50 137.6ms
multivector
Performance L4 b1 c16
Corpus 33.2K tok/s
Corpus p50 107.6ms
Query 33.8K tok/s
Query p50 137.6ms
bge_m3_flag
Quality
map at 10 0.7798
mrr at 10 0.7798
ndcg at 10 0.8066
default
Quality
map at 10 0.7798
mrr at 10 0.7798
ndcg at 10 0.8066