🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

We have the new way to build #enterpriseai with RAG 2.0. Our CTO Amanpreet Singh will be sharing how we accelerate #AI training workloads with @GoogleCloudNext tech. Join the discussion on April 10 → g.co/cloudnext

💼 I’ve joined Contextual AI as a research intern. I’ll be working on topics in AI alignment and retrieval-augmented generation. Let’s fry some GPUs!

Aligning Diffusion Models by Optimizing Human Utility We present Diffusion-KTO, a novel approach for aligning text-to-image diffusion models by formulating the alignment objective as the maximization of expected human utility. Since this objective applies to each

At #ICML2024, one may find our spotlight work on KTO Tuesday July 23 @ Hall C 4-9 #1204 :) Come chat with Kawin Ethayarajh and I about pluralistic model alignment and preference optimization!

Today we're releasing Not Diamond… The world’s most powerful AI model router. Not Diamond maximizes LLM output quality by automatically recommending the best LLM on every request at lower cost and latency. And it takes <5m to set up. Watch this to see how to start using it:

Launching the 1st Arena for Embedding Models: MTEB Arena🏟️ Vote @ hf.co/spaces/mteb/ar… ⚔️ 15 Models: OpenAI Google cohere Voyage AI Jina AI Salesforce AI Research Nomic AI E5 GritLM BGE.. 3 Tasks: Retrieval/Clustering/STS Deep dive with me on embeddings & the arena👇 🧵1/13

Embedding Models are key for building state-of-the-art Enterprise AI, but which model serves you best? Find out & vote in the Arena. 👇

This is the last chatbot you’ll ever need. Yesterday, Mckay Wrigley built an oss Not Diamond-powered chat app. We loved it. So today we’re releasing a hosted version. Get the best LLM on every message and hyper-personalize routing to your preferences with feedback. Watch how:

At Snowflake, we're making it easier than ever for customers to implement RAG while also enabling strict governance and privacy controls. That’s why we’re investing in Contextual AI—a powerful, end-to-end RAG solution that delivers highly accurate responses. With Contextual AI

Super excited to announce that we've raised our Series A! I'm very proud of what we've achieved as a team over the past year, and we're just getting started. 🚀🚀

Aligning Language Models with preferences leads to stronger and safer models (GPT3 → ChatGPT). However, preferences (RLHF) contain irrelevant signals, and alignment objectives (e.g. DPO) can actually hurt model performance. We tackle both, leading to a ~2x performance boost.

Enterprise AI systems need to be precisely aligned for each use case. We found that conventional alignment methods are underspecified, making this challenging. Today, we share solutions that tackle both alignment data and algorithms, resulting in a ~2x performance boost.

Looking for better LLM alignment techniques? With APO (Anchored Preference Optimization) ⚓️ and CLAIR (Contrastive Revisions) 🪞, you can precisely align your LLMs for a 2x performance boost! Check out this thread by the amazing Karel D’Oosterlinck for more details ⬇️

New promising RLHF and synthetic data generation method! 👀 APO and CLAIR improve the performance of Llama 3.1 8B by 7.45% on MixEval Hard, outperforming OpenAI GPT-4o mini! 🤯 Contrastive Learning from AI Revision (CLAIR): CLAIR is a synthetic preference dataset method where

Thank you NVIDIA for your ongoing partnership and covering us in your blog today. Contextual AI is proud to work with NVIDIA to develop and bring the next generation of LLMs, powered by RAG 2.0, to market. If you're building a production-ready enterprise use case using RAG,

OLMoE Open Mixture-of-Experts Language Models paper page: huggingface.co/papers/2409.02… We introduce OLMoE, a fully open, state-of-the-art language model leveraging sparse Mixture-of-Experts (MoE). OLMoE-1B-7B has 7 billion (B) parameters but uses only 1B per input token. We

We’re proud to share our latest research, led by our own Niklas Muennighoff and in partnership with Ai2: Introducing OLMoE, a best-in-class fully open source mixture-of-experts (MoE) language model with 1B active parameters that beats comparable LLMs and rivals many larger

Do you want to select great LLM pretraining data but don’t have 1000 H100s for a ton of mixture experiments? What about a method that requires none of your own training, matches the best known existing method, and has some nice theory? New preprint: Perplexity Correlations