EngineSamplerLatch class final
LiteRT-LM builds an engine's sampler at its first generation and keeps it for every later conversation (google-ai-edge/LiteRT-LM#2080), so a later session that asks for another sampler does not get it. This records the first and says what a later generation should report.
A larger topK later is covered too. LiteRT-LM 0.19 refuses a GPU
sampler k above the one the sampler was created with (upstream
6b799ae, TopKGpuSampler::CheckTopKWithinMax), but only in
UpdateConfig, whose one caller (the MTP drafter's
UpdateOrRecreateSampler) creates a new sampler for a larger k first;
the main executor never updates its sampler after the first generation
(InitializeSampler returns early). So a later session's larger topK
meets no native error: it is not applied, and this latch says so.
Constructors
Properties
- hashCode → int
-
The hash code for this object.
no setterinherited
- runtimeType → Type
-
A representation of the runtime type of the object.
no setterinherited
Methods
-
noSuchMethod(
Invocation invocation) → dynamic -
Invoked when a nonexistent method or property is accessed.
inherited
-
onGeneration(
Sampler sampling, {required bool explicit}) → ({String message, bool warn})? -
Call on every generation with its conversation's
samplingand whether the caller set any of it. Returns null when there is nothing to say; otherwise the message, withwarntrue for a release warning — once per engine, when the caller chose either sampler — and false for the verbose log, when both are defaults (a Qwen session with thinking on after one with it off). -
reset(
) → void - Forgets the engine's sampler: a new engine builds its own.
-
toString(
) → String -
A string representation of this object.
inherited
Operators
-
operator ==(
Object other) → bool -
The equality operator.
inherited