Jina AI

Better results scale with compute—more on learning, more on search. A good pretrained model takes you far, but test-time compute takes you further. It's important to recognize this new paradigm of scaling test-time compute, even for embedding models. https://lnkd.in/eFVArgjb

Comparing how long-context embedding models perform with different chunking strategies to find the optimal approach for your needs. https://lnkd.in/epJcKVg6

You use our embedding models to do what? This might be one of the most "out-of-domain" applications of embeddings we’ve seen at EMNLP 2024. https://lnkd.in/eanD3fCR

Jina-CLIP v2, a 0.9B multimodal embedding model with multilingual support of 89 languages, high image resolution at 512x512, and Matryoshka representations. https://lnkd.in/epHA8txA

Is Meta-Prompt the new norm for API specs? Feed it to LLMs and generate integration code that reliably integrates Jina's APIs, saving you from the usual trial-and-error process. https://lnkd.in/e5St-mcm

Search/acc! ⏩ Probably the hottest BoF session at #EMNLP2024! Nearly 100 researchers packed the room for 12 back-to-back talks on search foundation models - covering everything from code embeddings, distilled rerankers, ColPali, ColBERT, late chunking & smaller LMs. Killer lineup featuring Hans Ole Hatzel, Revanth Gangi Reddy, Björn Deiseroth, Rohan Jha, Manuel Faysse, Zhichao Xu, Jonghyun ­Song, Nayoung Choi, Rui Meng and many others! Big thanks to everyone who came out and showed such enthusiasm for search foundation models! We are looking forward to seeing you at EMNLP2025!

v3 + Late Chunking = cross-lingual positional alignment? In this Colab notebook, we show cross-lingual retrieval using jina-embeddings-v3, on different translations of "Alice's Adventures in Wonderland" Chapter 1. Two interesting observations: - The cross-lingual matches are remarkably accurate. - The matched segments across these different languages appear at roughly at the same position in their respective versions, showing a good alignment. This video explains why. Try it yourself - it runs smoothly on Google Colab's free T4 tier: https://lnkd.in/eu7CrX5A

At EMNLP 2024 Miami? Join us for a Birds of a Feather session focusing on embeddings, rerankers, and small LMs for better search. https://lnkd.in/eAVvryCS

Learn how Jina-CLIP enhances OpenAI's CLIP with better retrieval accuracy and more diverse results through unified text-image embeddings. https://lnkd.in/erq-fbPG

We trained three small language models to better segment long documents into chunks, and here are the key lessons we learned. https://lnkd.in/eQwnWzHw