#jina — Public Fediverse posts

2 new #jina models have entered the embedding arena — v5 multilingual:
* small: 1024 dim, 32K context
* nano: 768 dim, 8K context
both support matryoshka dimension truncation (32+) and are at the top of current benchmarks — especially for their parameter size
publicly accessible (non-commercial)

full announcement blog post: https://jina.ai/news/jina-embeddings-v5-text-distilling-4b-quality-into-sub-1b-multilingual-embeddings/
hugging face: https://huggingface.co/collections/jinaai/jina-embeddings-v5-text
paper for even more details: https://arxiv.org/abs/2602.15547

#VoyageAI introduces voyage-context-3, a contextualized chunk #embedding #llm that captures both chunk details and full document context 🔍 #ai

🔄 Outperforms #OpenAI-v3-large by 14.24% on chunk-level and 12.56% on document-level retrieval tasks

📊 Beats #Cohere-v4 by 7.89% and 5.64% respectively, and #Jina-v3 late chunking by 23.66% and 6.76%

🛠️ Drop-in replacement for standard embeddings without requiring downstream workflow changes

🧵👇

However, the distances were always similar without really getting any significant differences or ranking. So, I improved the embeddings with context (author notes from the account), tags, etc. I also tried different #jina models (base, small, large), with different max-tokens configuration. But the embeddings seems to be ok, the clusters were mostly random, but after some try and error I got some configuration I like it. So, what could be bad?

@raccoonbits Because my work I have been involved in other "AI" #LLM #RAG projects, and once I found #jina I thought it would be fun to integrate it using embeddings.

@SmartmanApps @dansup You are welcome to join testing my algo with #jina!

#Jina

Chinese spy balloon gathered intelligence from sensitive U.S. military sites, despite U.S. efforts to block it https://www.nbcnews.com/politics/national-security/china-spy-balloon-collected-intelligence-us-military-bases-rcna77155