SERVICE PROVIDER WORKBENCH v0.31

MaaS Training &
Fine-tuning Sizer

Minimum memory capacity and measured workload capacity — separate, explicit calculations.
Turkish / English interface · Help and examples follow the selected language

Training first. FULL training, LoRA and QLoRA share one memory and topology model. The optional MaaS overlay evaluates serving as a separate phase or a dedicated pool. Hover, focus or tap any yellow ? for help in the selected language.
Starting calculation…
06

Optimization profile & preview

Choose a goal and compare the current and proposed plan. Auto resolves both selected phases in one proposal. Training and MaaS profiles target their own phase and preserve any already runnable phase. Shared-pool changes are checked against both phases. Method, precision and workload are preserved. Explicit MaaS profiles propose Auto HA units; their changes are listed before Apply. The HA profile also enables physical-unit protection. Auto preserves the manual HA budget. Apply is available only when the selected goal is met.

Maximum profiles use the current pool when additional capacity is off, or the maximum node budget when it is on. In concurrent mode, the budget applies to the selected phase's pool.

Profile changes only take effect when applied.

Measured throughput optional · bind to the current recipe