Weights & Biases

Today we announced that we are being acquired by CoreWeave, the AI Hyperscaler. We could not be prouder or more excited to join forces with this team. We started Weights & Biases because we believed deeply in the power of AI to improve the world. It was and is our life’s mission to support the people out on the frontier pushing the research forward by building great software. CoreWeave has built complementary infrastructure and software for the same researchers and has an equally strong reputation for high quality service. We could only join forces with someone committed to continuing to deliver great products to our customers. And we’ve felt a shared deep respect for our customer base and a familiar intensity around moving quickly to support them. We look forward to collaborating with the CoreWeave team to build out an end-to-end platform for the world’s leading AI labs and enterprises to build, tune and deploy AI applications. https://lnkd.in/eDwMVaK4

?? Congrats to CoreWeave on the IPO! Turning raw compute into real momentum isn’t easy—today’s debut reflects just how far they’ve pushed the edge of what’s possible in AI infra.

Ready to scale AI enterprise-wide? Join us at #GoogleCloudNext! Our CEO and Co-Founder, Lukas Biewald will share proven strategies for deploying production-ready AI, managing complex rollouts, and driving measurable ROI. Plus, hear from Cresta and Glean on taking AI beyond pilots. Don’t get left behind—register now → g.co/cloudnext

?? Big news! Weights & Biases is recognized as an Emerging Leader in the 2025 Gartner?? Gartner Innovation Guide for Generative AI Technologies. ?? See here: https://lnkd.in/gWdkNHfx

SambaNova Systems is setting a new standard in LLM evaluation—combining custom AI hardware with powerful observability using W&B Weave. Their proprietary chips (RDUs) hit 198 tokens/sec on DeepSeek-R1 671B, achieving 3x the speed and 5x the efficiency of top GPUs—all from a single rack. That’s possible thanks to a unique 3-tier memory architecture enabling 100x faster model-switching. Hundreds of models stay in memory, making agentic AI truly scalable. But performance without visibility is a dead end. That’s where W&B Weave steps in—with full observability across inference performance and LLM operations. SambaNova uses W&B Weave to: ? Benchmark inference speed and latency ?? Track token usage and completion metrics ?? Analyze cost vs. performance across models ?? Score LLM outputs using an in-house LLM-as-a-judge evaluator ?? Evaluate both single and multi-step RAG pipelines ?? Log results directly into W&B dashboards for instant visualization They packaged all of this into the open-source SambaNova Evaluation Jumpstart Kit—built on W&B Weave, designed for fast, automated benchmarking across any LLM. Drop in a CSV or PDF. Hit “Evaluate.” Get answers—fast. Ready to see which model gives you the best tradeoff between speed, cost, and accuracy? Try their Evaluation Kit and optimize with confidence: https://lnkd.in/gEa7YmJt Learn more about how SambaNova Systems uses W&B Weave here: https://lnkd.in/g2sE3hge

How is AI reshaping military technology? ?? In the latest episode of Gradient Dissent, our host, Co-Founder and CEO Lukas Biewald, sits down with Captain Jon Haase (Jon J. H.) from the UNITED STATES DEPARTMENT OF THE NAVY for an insightful conversation on the real-world integration of AI and autonomous systems in defense operations. They discuss practical applications such as autonomous underwater vehicles detecting mines, multi-agent battlefield autonomy, and the critical importance of cybersecurity. Captain Jon Haase also shares thoughtful insights into the ethical considerations surrounding AI, particularly in lethal systems, reinforcing why human oversight remains central to warfighting decisions. Key highlights include: ?? Real-world applications of AI in mission-critical military operations ?? Challenges and innovations in underwater autonomy ?? Multi-agent collaboration and battlefield coordination ?? Cybersecurity considerations in autonomous systems ?? The emerging role of advanced language models like Gemini and Claude in defense This episode is essential listening for anyone interested in military technology, AI ethics, and the future of autonomous systems. ??? Catch the full conversation here:?https://lnkd.in/gy6PdRaf

Our new integration with CrewAI is now live in W&B Weave! ?? Debugging multi-agent AI systems has always been challenging—but not anymore. With this powerful integration, you can now automatically track every detail, including agents, tasks, LLM interactions, conditional routers, latency, and costs—all consolidated in one intuitive platform. This simplifies the entire debugging and monitoring process, making AI development faster, more transparent, and more effective. Get started here: https://lnkd.in/gbUqc5Z6 Later today, we’re teaming up with CrewAI to walk you through how you can use W&B Weave for AI agent evaluation & deployment. Kicks off at 8AM PT and will be recorded. RSVP here?? https://lnkd.in/gFSKBTxx Let's simplify and accelerate the future of AI agents together!

Our Weights & Biases 'AI Innovators meetup' is tomorrow (Wednesday, March 26) in Central London at 6pm! If you are interested in seeing how real-world applications of?#AI?are scaled to bring new products to life, then grab your ticket! Attend to hear from our customer,?Captur, who is pioneering the next generation of computer vision via edge AI with algorithms that can run compute, generate responses, and interact with users all on-device. Sarah Carthy?will be demoing the platform that Captur uses to build their AI models with confidence. The evening will wrap up with Q&A and drinks reception. Sign up to reserve a spot here ???https://lnkd.in/gvz-DvYQ?

We are here at NVIDIA GTC, where our CEO and Co-Founder Lukas Biewald took the stage yesterday to highlight what it really takes to turn GenAI prototypes into fully productionized solutions. The key message? It’s easy to demo AI, but the real challenge is ensuring it scales reliably in the real world. ? Evaluation is Essential: A solid framework for measuring performance and outcomes helps teams iterate quickly. ? User Feedback: Real-world insights are crucial—listen to your users early and often. ? Rapid Iteration: The best teams keep refining their prompts, data, and tooling to deliver consistent results. ? Reliable Tooling: Building with production-grade tools for versioning, monitoring, and collaboration helps avoid common GenAI pitfalls. We’re excited to share these lessons with you in person. If you’re at #GTC25, swing by our booth (#1336) to pick up some swag, talk shop with fellow developers, and see demos of how we’re putting these principles into practice. We’d love to say hello and hear about your own AI projects!

Finally got around to record and edit the talk I gave at AI Engineer summit NY a month ago. I wanted to explore whether reasoning models make better LLM judges. I had a lot of fun researching and giving this talk. Many asked for a recording / slides - YT link in comment