OptionalbatchMaximum ordinary continuous-batch rows.
OptionalcontextContext tokens the plan reserves.
OptionalkvThe composition's KV scheme and paging, which the runtime's caches compose (unset: unquantized, unpaged).
OptionallibraryNative loader library override.
OptionalmachineTest/diagnostic override; normal callers use physical RAM.
OptionalmaxGenerated-token allowance within contextTokens.
OptionalmemoryWhole-process ceiling in bytes.
OptionalnativeThe checkpoint-native draft head, on by default where the checkpoint has one.
Optionalpaged
Options every planned runtime understands; a runtime ignores what it has no use for. Unset values take the runtime's own preset.