workloads

Workload cards

Start from the request shape, then choose a qualified model and rig. These cards give the default direction; serving owns the complete API and operations contract.

• Interactive agents
• Long context and prefix reuse
• Batch throughput
• Multimodal
• Large models on multiple GPUs