naver/v-splade-quality
Primitive: /encode · Encode ·
ModernVBERT
V-SPLADE is Naver's visual SPLADE model: sparse lexical embeddings over document images and text on a ModernVBERT backbone, bringing interpretable SPLADE-style retrieval to visual documents
MultimodalSparse
Overview
Hardware: — drives latency, throughput & cost
| Size | 330M params |
|---|---|
| Tasks | /encode |
| License | apache-2.0 |
| Languages | en |
| Latency | — |
| Throughput | — |
| Cost | — /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: 50,368 |
| Max sequence length | 7,999 |
| Inputs | text · image |
Benchmarks
Vidore3ComputerScienceRetrieval
Visual document retrieval on computer science papers and slides
Quality
ndcg at 10 0.5839
map at 10 0.4565
mrr at 10 0.7488
Vidore3FinanceEnRetrieval
Visual document retrieval on financial reports
Quality
ndcg at 10 0.4651
map at 10 0.3504
mrr at 10 0.5965
Vidore3HrRetrieval
Visual document retrieval on HR-related documents
Quality
ndcg at 10 0.4481
map at 10 0.3195
mrr at 10 0.5624
Vidore3PharmaceuticalsRetrieval
Visual document retrieval on pharmaceutical documents
Quality
ndcg at 10 0.5324
map at 10 0.4196
mrr at 10 0.6386
Compare (0)Compare →