Connecting foundation models to physical systems —
from serverless GPU infrastructure to robots that act in the world.
Running VLA (Vision-Language-Action) model training on serverless GPU:
- SmolVLA fine-tuning on Norma Core — fine-tuning HuggingFace's SmolVLA foundation model for a physical robot arm using Nebius serverless jobs
- LeRobot fine-tune job — serverless training pipeline for LeRobot policies, from dataset to deployed model
Making real ML workloads run without managing infrastructure:
- Training, inference, and agentic workloads on serverless GPU
- POC → production, faster — based on what actually breaks in practice
→ Try it: console.nebius.com/serverless/job
→ Cookbook (all examples): github.com/mnrozhkov/serverless-cookbook
| Coding | AI & ML | Infra | Robotics & HW |
|---|---|---|---|
If you're working on embodied AI, VLA models, or GPU infrastructure for physical systems — let's talk.
📬 LinkedIn · 🐦 X · 🤖 ML REPA Community