CoreWeave targets GPU utilization with Forge for continuous AI post-training

CoreWeave announced Forge, a platform connecting model deployment, evaluation and improvement, plus a preview of RL Rollouts for repeated training cycles. With You.com and Nvidia, post-training Nemotron 3.5 Lightning with web search took eight hours.

• Forge links deployment, evaluation and model improvement in one loop
• RL Rollouts preview speeds weight sync via nearby peers
• AI Object Storage adds cross-region writes for post-training jobs
• Nemotron 3.5 Lightning post-trained in 8 hours with You.com search