lloyal-agents API Reference
    Preparing search index...

    Type Alias KvCacheType

    KvCacheType:
        | "f32"
        | "f16"
        | "bf16"
        | "q8_0"
        | "q4_0"
        | "q4_1"
        | "iq4_nl"
        | "q5_0"
        | "q5_1"

    Supported KV cache quantization types

    Matches llama.cpp CLI -ctk / -ctv flags. Lower precision = less GPU memory, slight quality tradeoff.