Featured Post


Fine-Tuning Multimodal Ecommerce Search on Over 100 Million Products

While working with an e-commerce retailer with a large product catalogue of over 110 million items, we put together metrics on performance using this real world data. We constructed an index in Marqo using a ViT-B-32 model that we had fine-tuned on over 1.5 billion (real) query interactions using Generalised Contrastive Learning (GCL). The model was fine-tuned to match queries to a combination of title, description, and image. It also used signals from live search to learn a ranking beyond what can be obtained using only binary relationships.

Read Article

Latest Posts

Request Your Marqo Demo Today!

Discover the transformative capabilities of Marqo. Schedule your personalized demo now and elevate your workflow.

Book Demo