ThinkingBudget class
Limits the generated tokens within a model's reasoning block.
This maps to llama.cpp's reasoning-budget sampler. A budget of zero closes the next reasoning block immediately; a positive value permits that many generated reasoning tokens before the sampler forces endTag. llama.cpp accepts values from zero through 2,147,483,647.
LlamaEngine.create resolves omitted startTag and endTag from the rendered chat template. Callers using raw generation should supply both tags explicitly because models do not share a universal reasoning syntax.
Constructors
- ThinkingBudget({required int maxTokens, String? startTag, String? endTag, String? forcedMessage})
-
Creates a thinking-token budget.
const
Properties
- endTag → String?
-
Reasoning-block closing delimiter forced when maxTokens is exhausted.
final
- forcedMessage → String?
-
Optional text forced immediately before endTag after exhaustion.
final
- hashCode → int
-
The hash code for this object.
no setterinherited
- maxTokens → int
-
Maximum generated tokens allowed inside each reasoning block.
final
- runtimeType → Type
-
A representation of the runtime type of the object.
no setterinherited
- startTag → String?
-
Reasoning-block opening delimiter.
final
Methods
-
noSuchMethod(
Invocation invocation) → dynamic -
Invoked when a nonexistent method or property is accessed.
inherited
-
toString(
) → String -
A string representation of this object.
inherited
Operators
-
operator ==(
Object other) → bool -
The equality operator.
inherited