Enum KVCachePrecisionMode

Namespace

AiDotNet.Configuration

Assembly

AiDotNet.dll

Controls the numeric precision of KV-cache storage.

public enum KVCachePrecisionMode

Fields

Auto = 0

Select an industry-standard default.

Uses FP16 when KV-cache is enabled and the numeric type supports conversion; otherwise falls back to FP32.

Float16 = 1

Store KV-cache in FP16 (half precision) to reduce memory use.

Float32 = 2

Store KV-cache in FP32 (single precision) for maximal numerical fidelity.