Why did we open-source our inference engine? Read the post

← Catalog

laion/CLIP-ViT-H-14-laion2B-s32B-b79K

Open comparison →

Primitive: /encode · Encode · CLIP

1. Model Details 3. Training Details 4. Evaluation 5. Acknowledgements 6. Citation 7. How To Get Started With the Model

MultimodalDense

Overview

Hardware: — drives latency, throughput & cost

Size986M params
Tasks /encode
Licensemit
Latency353 ms
Throughput438 tok/s
Cost$0.508 /1M tok

Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.

Embedding

Output typesDense
Dimensionsdense: 1,024
Max sequence length77
Inputstext · image

Benchmarks

Flickr30kI2TRetrieval

general retrieval en

Image-to-text retrieval: retrieve captions from images

Corpus: 31,783 Queries: 1,000
Quality
ndcg at 10 0.8624
map at 10 0.7856
mrr at 10 0.9488
Performance L4-SPOT b1 c8
Corpus 181 tok/s
Corpus p50 533.4ms
Query 10.2 img/s
Query p50 625.9ms
Performance L4 b1 c16
Corpus 695 tok/s
Corpus p50 172.6ms
Query 10.1 mpix/s
Query p50 141.9ms
Reference →

Open source inference for agents

Open-source inference for the models behind your agents. Run it yourself, or let us run it for you.

Github 2.1K

Contact us

Tell us about your use case and we'll get back to you shortly.

Apply for an inference grant

Free capacity on our hosted cluster for selected projects. Tell us what you run and we reply by email.