Luminide

We are beyond thrilled to share our first flagship models, Rnj-1 base and instruct 8B parameter models. Rnj-1 is the culmination of 10 months of hard work by a phenomenal team, dedicated to advancing American SOTA OSS AI. Lots of wins with Rnj-1. 1. SWE bench performance close to GPT 4o. 2. Tool use outperforming all comparable open source models. 3. Mathematical reasoning (AIME’25) nearly at par with GPT OSS MoE 20B. ….

[1/5] 🚀 Meet Essential-Web v1.0, a 24-trillion-token pre-training dataset with rich metadata built to effortlessly curate high-performing datasets across domains and use cases!

Rethinking Reflection in Pre-Training This is one of the more interesting papers I read this week. It argues that reflection emerges during pre-training. It introduces adversarial reasoning tasks to show that self-reflection and correction capabilities steadily improve as compute increases, even in the absence of supervised post-training. Key contributions: • Propose two kinds of reflection: - Situational-reflection: The model corrects errors in others’ reasoning. - Self-reflection: The model corrects its own prior reasoning. • Build six adversarial datasets (GSM8K, TriviaQA, CruxEval, BBH) to test reflection across math, coding, logic, and knowledge domains. On GSM8K-Platinum, explicit reflection rates grow from ~10% to >60% with increasing pre-training tokens. • Demonstrate that simple triggers like “Wait,” reliably induce reflection. • Evaluate 40 OLMo-2 and Qwen2.5 checkpoints, finding a strong correlation between pre-training compute and both accuracy and reflection rate. Why it matters: • Reflection is a precursor to reasoning and can develop before RLHF or test-time decoding strategies. • Implication: We can instill advanced reasoning traits with better pre-training data and scale, rather than relying entirely on post-training tricks. • They also show a trade-off: more training compute reduces the need for expensive test-time compute like long CoT traces.

Reinforcement learning has shown success in eliciting reflection from LLMs, but what if this capability actually manifests earlier in pre-training? We investigated this question and our results are surprising 👇 [1/4]

More examples are available at: github.com/akridata-ai/ak… Read more about Akridata at: akridata.ai/data-explorer/

So happy to hear this news and very excited for my friends and former (and current) colleagues, what this means for the open source LLM eco-system and an amazing outcome as an angel investor. Big congrats @NaveenGRao and @hanlintang and the entire @MosaicML team!

Incredibly grateful to our community and customers who have made this journey possible. Amid the breathless twitter overhype and head-spinning pace of change in this field, I have always felt grounded and focused on "just make our customers happy" and tune out the noise.

🔋Build and deploy LLM apps even faster with Log10! - We've added several examples covering use-cases relevant to embedding LLM powered applications in DevTools, CRMs and other SaaS products - Agents take the grunt work out of prompt engineering - Iterate faster with Playgrounds and hosted Applications - No API key required to get started!

I'll be talking about Computer Vision tomorrow. It's free. You can come and ask any questions you'd like. Here is the link: us06web.zoom.us/webinar/regist…

📣 Join us on March 30th where we will demonstrate how the latest AI platforms for visual data can help you take a data-centric approach to preparing data for training Computer Vision models. Register now! us06web.zoom.us/webinar/regist…

🚨 Attention all AI/ML professionals! 🚨 Join us for a 30-minute webinar on 3/30 at 1:00PM PT with @Alex_AkriData, Principal AI/ML Engineer at Akridata, as he showcases the latest AI platforms for visual data. Register now! us06web.zoom.us/webinar/regist…

Thanks, Mario! #AI #ML #DataScience #AgTech @Kaggle linkedin.com/posts/caesarma…

Mark your calendars! nvidia.com/gtc/session-ca… @NVIDIAGTC @NVIDIADRIVE @NVIDIAAI @NVIDIA

Excited to see @huggingface ramping up its competitions effort with a new #ComputerVision competition. The big question is how this will compare with @kaggle. And with everything else -- papers, models, datasets, deployment, spaces, etc. -- will HF now become the "home of ML" ?

Papers, models, datasets, autotrain, inference, deployment, competitons, spaces...and more!!!! can we call Hugging Face the home of machine learning now???!!!! 🤗🚀🎉

@dk21 congrats!

Earned my 5th gold medal and Kaggle GM rank - want to lough and cry at the same time... felt like I was playing in the WC finals today, being on a great team and fighting to the end - the gold sub finished running 15 mins before deadline 😅😂❤️

📣 Join us on March 30th for a webinar where we will demonstrate how the latest AI platforms for visual data can help you take a data-centric approach to preparing data for training Computer Vision models. Register now! us06web.zoom.us/webinar/regist…

We're making it easier than ever to explore your #datasets! Data Explorer allows you to curate, clean, and visualize your data so that you can make sure your algorithms and #models are based on strong foundations. akridata.ai/blog/image-dat…

It's here... openai.com/research/gpt-4