CoreWeave ARENA is where teams validate AI workloads end to end before committing to production. In this demo, an agent that builds marketing distribution lists from user behavior runs on an open-weight model served through CoreWeave Serverless Inference. Weights & Biases Weave collects the agent’s traces and uses signals to flag low-quality responses and mismatches between the user request and the SQL the agent generates. See how:
- CoreWeave Serverless RL runs the reinforcement learning job with instant GPU access and no provisioning
- A shared workspace tracks run progress and reward scores for model checkpoints
- Before-and-after evaluations compare model performance across individual metrics
