Valohai

Is AMD's MI300X GPU the best pick for LLM inference on a single GPU ❓ As our mission is to offer the leading MLOps platform, we're constantly engaged in boundary-pushing R&D work that involves testing and comparing the latest hardware and software. Most of this work never sees the light of day. But this time, we're confident that we've come across something so awesome that we can't keep it under the covers. 👇 We've conducted benchmarks of GPU performance for LLM inference on a single GPU, comparing Nvidia's popular H100 and AMD's new MI300X GPU. We found that AMD's MI300X GPU can be a better fit for handling large models on a single GPU due to its larger memory and higher memory bandwidth. Take a deep dive with us and learn about the impact on AI hardware performance and model capabilities in our blog. Link in the comments 👇

Meet the team behind Valohai in person and hear some next-level keynotes 📢 We're coming to three industry events in the US and Sweden this November: 1️⃣ MLOps World by Toronto Machine Learning Society (TMLS) 🇺🇸 Austin, US 📅 November 7-8 Our CEO, Eero Laaksonen will talk about how to avoid the common pitfalls when scaling your ML operations based on his insights from over a thousand ML teams. 2️⃣ MLOps Community meetup 🇸🇪 Stockholm, Sweden 📅 November 7 Our Head of Product, Tarek Oraby will give a talk on how to automate the mechanisms for AI governance. Many thanks to Patrick Couch for making this happen! Sign up here before the seats run out: https://lnkd.in/dv8KK3rS 3️⃣ AI in Healthcare & Pharma Summit by RE•WORK 🇺🇸 Boston, US 📅 November 13-14 We'll announce the talk very soon. (Hint: Advanced medical imaging and the complex ML infrastructure behind it.) Can't join these events? We could still meet you in Austin, Stockholm, and Boston. Don't hesitate to drop us a line at: hello@valohai.com 

Let’s take a closer look at Valohai’s new Model Hub 🔎 In short, it’s a central control plane that gives you the easiest way to automate model lifecycle management: 1️⃣ Overview all your models in one place Model Hub’s front page gives you a holistic view of all your models for specific projects or on the organizational level. From here, you can navigate further and learn about every model in more detail. 2️⃣ Get an in-depth look at every model Valohai automatically keeps track of all model versions and their approval status. In addition, you can document models by adding custom tags and descriptions to improve collaboration and compliance even further. If you’re an organization admin, you can also manage access control to the specific models 🔐 3️⃣ Trace the entire lineage of each model version Even before the release of the Model Hub, Valohai automatically tracked all assets of each model version, such as artifacts, datasets, sources, and metrics. The Model Hub leverages this functionality to help you take the guesswork out of tracing model lineage and ensuring that all the results are reproducible. 4️⃣ Automate complex workflows Model Hub supports triggers that can be used to automate workflows, such as deploying a new model version while revoking the previous version. But we're only scratching the surface. The Model Hub comes packed with many more advanced features. Learn more and give it a try at: https://hubs.ly/Q02RsydK0

🥁 Introducing a new major addition to the Valohai MLOps platform: the Model Hub 🥁 We’ve built the Model Hub to give machine learning teams the easiest way to manage and track model versions across their entire lifecycle. It comes with advanced features such as automated versioning, lineage tracking, performance comparison, workflow automation, access control, and many more. Learn more and get started at: https://hubs.ly/Q02R2fPC0

We'll be publishing 3 new stories for you over the next 3 weeks 🙌 Which one do you look forward to the most? If you don't want to miss these updates, here's a friendly nudge to subscribe to our newsletter at: https://hubs.ly/Q02PPZc30

We have 3 exciting stories to share with you over the next few weeks: 🔸 AI governance and the AI EU Act 🔸 A new key feature in the Valohai MLOps platform 🔸 How to build production pipelines for scheduled retraining an deployment If you'd love to get notified once these stories are out, subscribe to our MLOps-themed newsletter at: https://hubs.ly/Q02Ptvk50

Valohai's new Smart Instance Selection helps you choose machines where your training data is cached, saving time on data transfers. With this feature, you can optimize compute resources and increase iteration speed by clicking on a single check box in the UI. Learn more and get started at: https://hubs.ly/Q02P45tK0

Be among the first to test our new experimental feature, Smart Instance Selection. Here's how it works: 1️⃣ When enabled, Valohai will proactively analyze historical job data to identify instances with the highest cache hit rates. 2️⃣ When a new job is submitted, the platform will prioritize assigning it to an instance with cached data. 🆒 If no instances with cached data are available, the system will revert to the default first-in, first-out queueing behavior. Give it a try at: https://hubs.ly/Q02Nn3Xs0

Download speed: way too slow 🐌 Time remaining: infinity ♾️ Rinse and repeat 🧺 Or better not! With our new feature, you can avoid downloading the same training data sets over and over again. Here's how it works: After submitting a new job to Valohai, the platform will automatically select the machine that has the necessary data cached from previous runs. Learn more and get started at: https://hubs.ly/Q02Nn3j50

Our new integration with OVHcloud enables you to scale computational resources on-demand and without changing your ways of working. By pairing Valohai's agnostic MLOps platform with OVHcloud's secure environments, you can control costs while tracing and reproducing all runs. Learn more and get started at: https://hubs.ly/Q02Nh55z0