Class DeepJNova
java.lang.Object
io.github.kirstenali.deepj.nn.transformer.DecoderOnlyModel
io.github.kirstenali.deepj.nn.transformer.nova.DeepJNova
- All Implemented Interfaces:
CausalLM,Persistable,Trainable
A decoder with grouped-query attention, optional QK normalization, and tied embeddings.
-
Field Summary
Fields inherited from class io.github.kirstenali.deepj.nn.transformer.DecoderOnlyModel
lmHead, normF, stack, tokEmb -
Constructor Summary
ConstructorsConstructorDescriptionDeepJNova(DeepJNovaConfig config, long seed) DeepJNova(DeepJNovaConfig config, long seed, ParameterFactory parameters) DeepJNova(DeepJNovaConfig config, long seed, DType dtype) -
Method Summary
Methods inherited from class io.github.kirstenali.deepj.nn.transformer.DecoderOnlyModel
applyInitScale, embed, embed, embedBatch, embeddingParameters, forward, forwardBatch, load, metadata, openGeneration, openGeneration, promptFormat, save, setMetadata, setPromptFormat, supportsBatchedForward
-
Constructor Details
-
DeepJNova
-
DeepJNova
-
DeepJNova
-
-
Method Details
-
config
- Specified by:
configin classDecoderOnlyModel
-
gradClipNorm
public float gradClipNorm() -
supportsCachedGeneration
protected boolean supportsCachedGeneration()- Overrides:
supportsCachedGenerationin classDecoderOnlyModel
-
supportsRollingGeneration
protected boolean supportsRollingGeneration()- Overrides:
supportsRollingGenerationin classDecoderOnlyModel
-
parameters
- Specified by:
parametersin interfacePersistable- Specified by:
parametersin interfaceTrainable- Overrides:
parametersin classDecoderOnlyModel
-