jackboyla/glirel-large-v0
Primitive: /extract · Extract ·
DeBERTa
GLiREL is a Relation Extraction model capable of classifying unseen relations given the entities within a text. This builds upon the excelent work done by Urchade Zaratiana, Nadi Tomeh, Pierre Holat, Thierry Charnois on the GLiNER library which enables efficient zero-shot Named Entity Recognition.
Relations
Overview
Hardware: — drives latency, throughput & cost
| Size | 435M params |
|---|---|
| Tasks | /extract |
| License | — |
| Latency | 105 ms |
| Throughput | 7.3K tok/s |
| Cost | $0.030 /1M tok |
Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.
Extraction
| Output kinds | Relations |
|---|---|
| Inputs | text |
| Max sequence length | — |
Benchmarks
FewRel
Few-shot relation extraction from Wikipedia sentences
Corpus: 70,000 Queries: 70,000
Quality
f1 0.2639
precision 0.2397
recall 0.2934
Performance L4-SPOT b1 c16
Performance L4 b1 c16
Compare (0)Compare →