google/siglip-so400m-patch14-224
Primitive: /encode · Encode ·
SigLIP
SigLIP model pre-trained on WebLi at resolution 224x224. It was introduced in the paper Sigmoid Loss for Language Image Pre-Training by Zhai et al. and first released in this repository.
MultimodalDense
Overview
Hardware: — drives latency, throughput & cost
| Size | 877M params |
|---|---|
| Tasks | /encode |
| License | apache-2.0 |
| Latency | 284 ms |
| Throughput | 456 tok/s |
| Cost | $0.487 /1M tok |
Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.
Embedding
| Output types | Dense |
|---|---|
| Dimensions | dense: 1,152 |
| Max sequence length | 64 |
| Inputs | text · image |
Benchmarks
Flickr30kI2TRetrieval
Image-to-text retrieval: retrieve captions from images
Corpus: 31,783 Queries: 1,000
Quality
ndcg at 10 0.8383
map at 10 0.7481
mrr at 10 0.9353
Performance L4-SPOT b1 c8
Corpus 223 tok/s
Corpus p50 395.0ms
Query 11.5 img/s
Query p50 392.1ms
Performance L4 b1 c16
Corpus 689 tok/s
Corpus p50 173.8ms
Query 10.6 mpix/s
Query p50 135.9ms
Compare (0)Compare →