Post by Michael Johnston

🚀 𝗪𝗲’𝘃𝗲 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲𝗱 𝗮𝗱𝗼 — the framework behind our large-scale GenAI benchmarking efforts! If you’ve ever struggled with reproducibility, scaling, or managing thousands of computational experiments, ado (the accelerated discovery orchestrator) might be just what you need. It’s already accelerating our work in many domains including fine-tuning performance and geospatial inference pipeline performance. ado helps researchers and engineers: ✅ Share and reuse results across teams ✅ Track provenance and ensure reproducibility ✅ Scale seamlessly on Kubernetes or bare-metal (via #Ray) ✅ Leverage rigorous validation to minimize errors ✅ Run optimizations, grid searches, or configuration space refinement ✅ Easily add new operators (for driving explorations) and actuators (for executing experiments in specific domains) We’ve distilled our experience from running tens of thousands of experiments into a flexible, modular framework — and now it’s open for everyone to try. You can get started in minutes with our examples, including fine-tuning benchmarks that run right on your laptop. 🛠️ Try it out: https://github.com/ibm/ado 🔗 Learn more in our blog: https://lnkd.in/eUzZP4De We’d love to hear your thoughts on what does and doesn't work for your use-cases and how we can improve it! Big thanks to everyone who contributed to ado through code and discussions, past and present: Alessandro Pomponio, Vassilis Vassiliadis, Srikumar Venugopal, Christian Pinto, Michele Gazzetti, Daniele Lotito, Boris Lublinsky, Burkhard Ringlein, Christoph Hagleitner, James C. Sexton, Renato Cerqueira, Renato Maia Dr. Juan Bernabe Moreno, IBM Research #GenAI #Ray #AIDev #Bencmarking