Uses of Interface
ai.beestream.jitnn.RopeScaling
Packages that use RopeScaling
Package
Description
The jitNN API:
Tensor (n-dimensional arrays, lazy on TornadoVM
devices), reverse-mode differentiation (Tensor.backward(), Grad),
the functional operators of F, linear algebra (Linalg),
Fourier transforms (Fft), sparse matrices (SparseMatrix),
mixed precision (Autocast) and user-defined operators
(CustomOp).Complete models: ResNet, MobileNetV2, GPT-2, a Llama-style decoder, BERT, a VAE, DCGAN, U-Net and
character RNNs, and the embedding models word2vec, doc2vec and GloVe.
-
Uses of RopeScaling in ai.beestream.jitnn
Classes in ai.beestream.jitnn that implement RopeScalingModifier and TypeClassDescriptionstatic final recordPositions divided byfactor.static final recordLlama 3.1's rescaling; seeRopeScaling.llama3(double, double, double, int).Methods in ai.beestream.jitnn that return RopeScalingModifier and TypeMethodDescriptionstatic RopeScalingRopeScaling.linear(double factor) Positions divided byfactor(rope_type linear).static RopeScalingRopeScaling.llama3(double factor, double lowFreqFactor, double highFreqFactor, int originalContext) Llama 3.1's rescaling (rope_type llama3): frequencies with wavelengths above originalContext / lowFreqFactor are divided byfactor, those below originalContext / highFreqFactor are kept, and the band between is interpolated.Methods in ai.beestream.jitnn with parameters of type RopeScalingModifier and TypeMethodDescriptionstatic TensorF.ropeHeads(Tensor x, float base, RopeScaling scaling, int offset, boolean interleaved) F.ropeHeads(Tensor, float, int, boolean)with the frequencies rescaled for longer contexts (Llama 3.1's rope_scaling; null for none). -
Uses of RopeScaling in ai.beestream.jitnn.nn.models
Methods in ai.beestream.jitnn.nn.models with parameters of type RopeScalingModifier and TypeMethodDescriptionLlama.ropeScaling(RopeScaling scaling) Rescales the RoPE frequencies of every block (Llama 3.1's rope_scaling); null for none.