opensearch-project/opensearch-neural-sparse-encoding-v2-distill
Primitive: /encode · Encode ·
DistilBERT
The model should be selected considering search relevance, model inference and retrieval efficiency(FLOPS). We benchmark models' zero-shot performance on a subset of BEIR benchmark: TrecCovid,NFCorpus,NQ,HotpotQA,FiQA,ArguAna,Touche,DBPedia,SCIDOCS,FEVER,Climate FEVER,SciFact,Quora.
Overview
Hardware: — drives latency, throughput & cost
| Size | 67M params |
|---|---|
| Tasks | /encode |
| License | apache-2.0 |
| Languages | en |
| Latency | 63 ms |
| Throughput | 44.2K tok/s |
| Cost | $0.0050 /1M tok |
Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.
Embedding
| Output types | Sparse |
|---|---|
| Dimensions | sparse: 30,522 |
| Max sequence length | 512 |
| Inputs | text |
Benchmarks
CQADupstackPhysicsRetrieval
Duplicate question retrieval from StackExchange Physics
CosQA
Code search with natural language queries
FiQA2018
Financial opinion mining and question answering
LegalBenchConsumerContractsQA
Question answering on consumer contracts
NFCorpus
Biomedical literature search from NutritionFacts.org
SCIDOCS
Citation prediction, document classification, and recommendation for scientific papers
SciFact
Scientific claim verification using research literature
StackOverflowQA
Programming question answering from Stack Overflow