GPT class - gpt library

Constructors

config → GPTConfig: final
embedDrop → Dropout: final
encoder → TransformerEncoder: final
hashCode → int: The hash code for this object.
no setterinherited
posEmb → LearnedPositionalEmbedding: final
runtimeType → Type: A representation of the runtime type of the object.
no setterinherited
tokenEmb → Embedding: final
training ↔ bool: Whether this module is in training mode. Layers that behave differently between training and inference (e.g. Dropout) read this flag in their call method. Defaults to training mode.
getter/setter pairinherited
untiedHead → Linear?: Only populated when config.tieWeights == false. When tied, the output head is computed inline as h @ tokenEmb.weight.T.
final

call(Tensor tokens) → Tensor: Forward pass. tokens is either 1D [seqLen] (single sequence, output [seqLen, vocabSize]) or 2D [batch, seqLen] (batched, output [batch, seqLen, vocabSize]). A causal mask is applied inside the encoder. KV caching is not used on this path.
eval() → void: Put this module (and any registered submodules) into evaluation mode.
inherited
generate(List<double> prompt, {required int maxNewTokens, double temperature = 1.0, int? topK, Random? rng, bool useCache = true}) → List<double>: Autoregressive sampling. Returns the full sequence prompt + generated tokens as a Dart list of doubles (matches the float32 index convention used everywhere else). The model is put into eval mode for the duration of the call and restored on exit.
noSuchMethod(Invocation invocation) → dynamic: Invoked when a nonexistent method or property is accessed.
inherited
parameters() → List<Tensor>: Trainable tensors owned by this module (and its submodules).
override
submodules() → List<Module>: Submodules owned by this module. Subclasses that compose other modules should override this so train() / eval() propagate. Default: empty.
override
toString() → String: A string representation of this object.
inherited
train() → void: Put this module (and any registered submodules) into training mode.
inherited
zeroGrad() → void: Zero every parameter's gradient. Safe to call before each backward.
inherited