Table of Contents

Enum HibernationMode

Namespace
LMKit.Inference
Assembly
LM-Kit.NET.dll

Controls when the runtime hibernates an inference context at the end of a conversation turn. The mode is set on HibernationMode by the owning conversation (e.g. MultiTurnConversation, PdfChat) and consulted by the inference backend every time a usage lock is released.

public enum HibernationMode

Fields

Auto = 0

Default. At the end of a turn the runtime hibernates the context only when the device is over its cached-context memory budget (MaxCachedContextMemoryRatio); otherwise the context stays resident and is marked eligible for the runtime to reclaim later under memory pressure. This keeps hot sessions fast while letting the runtime free memory automatically when it runs short.

None = 1

No explicit hibernation. The runtime keeps the context resident in memory between turns; callers may still trigger hibernation manually via HibernateAsync(string).

Forced = 2

At the end of every conversation turn, the runtime schedules HibernateAsync(string) as a fire-and-forget background task: the caller does not wait for the KV-cache to land on disk, and the next turn rehydrates the context transparently from the on-disk image.

Share