GenConfig class

Generation parameters for ExecuTorchLLM.generate.

Fields mirror ExecuTorch's extension/llm/runner GenerationConfig exactly. Sampling is temperature-only — there is intentionally no top-p / top-k, because the upstream runner does not support them.

Constructors

GenConfig({int maxNewTokens = -1, int seqLen = -1, double temperature = 0.8, bool echo = false, bool ignoreEos = false, int numBos = 0, int numEos = 0})
Creates a generation config. Defaults match the upstream runner defaults.
const

Properties

echo → bool
Whether to echo the input prompt back at the start of the output stream.
final
hashCode → int
The hash code for this object.
no setterinherited
ignoreEos → bool
Whether to keep generating past the EOS token up to the token limit.
final
maxNewTokens → int
Maximum number of new tokens to generate.
final
numBos → int
Number of BOS tokens to prepend to the prompt.
final
numEos → int
Number of EOS tokens to append to the prompt.
final
runtimeType → Type
A representation of the runtime type of the object.
no setterinherited
seqLen → int
Maximum number of total tokens (prompt + generated).
final
temperature → double
Sampling temperature in [0, 1] (higher = more random). 0 is greedy (argmax); values outside the range fail the generation.
final

Methods

copyWith({int? maxNewTokens, int? seqLen, double? temperature, bool? echo, bool? ignoreEos, int? numBos, int? numEos}) → GenConfig
Returns a copy with the given fields replaced.
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