Interface IChatModelHost

Namespace

Qavren.Edge.Chat

Assembly

Qavren.Edge.Chat.Onnx.dll

The ref-counted chat-model host. One per AddOnnxChat registration.

public interface IChatModelHost

Properties

IsAcceptingTurns

False between MemoryPressure(Moderate) and the next Resumed.

Methods

AcquireAsync(CancellationToken)

Awaits IEdgeHost.EnsureStartedAsync, verifies provisioning, resolves the budget, loads on first call, and returns the exclusive lease.

Describe()

What the host last knew about the model, or null if it has never loaded one.

PreloadAsync(CancellationToken)

Everything AcquireAsync(CancellationToken) does except awaiting host startup, so an IEdgeStartupTask can drive a load without deadlocking on its own completion. Public on purpose: sub-project 2 documents this exact hole as unclosable from L1 because its equivalent is internal to another assembly. Sub-project 4's host and its warm-up task are in the same assembly, so sub-project 4 does not inherit it.

TerminateActiveGeneration()

Sets a volatile abort flag AND calls Generator.SetRuntimeOption("terminate_session", "1") on the live generator - the documented cooperative abort, and the only thing that can stop a multi-second prefill. Safe to call from a lifecycle observer on another thread; it is the one GenAI call sub-project 4 makes outside the turn gate, and section 20 records that as a stated risk.

UnloadAsync(CancellationToken)

Marks the model for drop and disposes it once the lease returns.