Interface TransformerConfig
- All Known Implementing Classes:
DeepSeekConfig,GPTConfig,LlamaConfig
public interface TransformerConfig
Common configuration fields shared by all decoder-only transformer models.
Implemented by GPTConfig and
LlamaConfig.
-
Method Summary
Modifier and TypeMethodDescriptionintdFF()intdModel()floatintintnHeads()intnLayers()static voidvalidateCommon(int vocabSize, int maxSeqLen, int dModel, int nHeads, int nLayers, int dFF, float gradClipNorm) Validates the fields common to all transformer configs.static voidvalidateRotaryHeadDimension(int dModel, int nHeads) Validates the even per-head width required by rotary embeddings.int
-
Method Details
-
vocabSize
int vocabSize() -
maxSeqLen
int maxSeqLen() -
dModel
int dModel() -
nHeads
int nHeads() -
nLayers
int nLayers() -
dFF
int dFF() -
gradClipNorm
float gradClipNorm() -
validateCommon
static void validateCommon(int vocabSize, int maxSeqLen, int dModel, int nHeads, int nLayers, int dFF, float gradClipNorm) Validates the fields common to all transformer configs. Call from each record's compact constructor. -
validateRotaryHeadDimension
static void validateRotaryHeadDimension(int dModel, int nHeads) Validates the even per-head width required by rotary embeddings.
-