Index

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z 
All Classes and Interfaces|All Packages|Constant Field Values|Serialized Form

A

abort() - Method in interface ai.beestream.jitnn.dist.Collective
Releases the ranks waiting in a collective after another rank failed.
abort() - Method in class ai.beestream.jitnn.dist.HostCollective
Releases every rank waiting in a collective with an error (called when a rank fails).
abort() - Method in class ai.beestream.jitnn.dist.NcclCollective
Releases ranks waiting in a collective after another rank failed; the communicator is unusable after it.
abs() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
abs() - Method in class ai.beestream.jitnn.Tensor
|x|, element-wise.
accept(double) - Method in interface ai.beestream.jitnn.onnx.Proto.DoubleSink
Receives one value.
accept(float) - Method in interface ai.beestream.jitnn.onnx.Proto.FloatSink
Receives one value.
accept(long) - Method in interface ai.beestream.jitnn.onnx.Proto.LongSink
Receives one value.
accuracy(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Fraction of rows whose argmax equals the label; runs now and returns a host value.
acos() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
acos() - Method in class ai.beestream.jitnn.Tensor
Element-wise arccosine.
acosh() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
acosh() - Method in class ai.beestream.jitnn.Tensor
Element-wise inverse hyperbolic cosine.
AdaDelta - Class in ai.beestream.jitnn.nn.optim
AdaDelta (Zeiler), PyTorch's semantics: decay rho, eps, learning rate (default 1), L2 weight decay.
AdaDelta(List, float, float, float, float) - Constructor for class ai.beestream.jitnn.nn.optim.AdaDelta
AdaDelta with decay rho, eps and L2 weight decay.
AdaGrad - Class in ai.beestream.jitnn.nn.optim
AdaGrad (Duchi et al.), PyTorch's semantics: learning-rate decay, L2 weight decay, initial accumulator.
AdaGrad(List, float, float, float, float, float) - Constructor for class ai.beestream.jitnn.nn.optim.AdaGrad
AdaGrad with learning-rate decay, L2 weight decay, the accumulator's starting value, and eps.
Adam - Class in ai.beestream.jitnn.nn.optim
Adam (Kingma and Ba) with L2 weight decay, or AdamW (decoupled decay) via Adam.adamW(List, float, float); with amsgrad, AMSGrad's running maximum of the second moment (Reddi et al.).
Adam(List, float) - Constructor for class ai.beestream.jitnn.nn.optim.Adam
Adam with PyTorch's defaults: betas 0.9 and 0.999, eps 1e-8, no weight decay.
Adam(List, float, float, float, float, float, boolean) - Constructor for class ai.beestream.jitnn.nn.optim.Adam
Adam with the given hyper-parameters; decoupled makes the weight decay AdamW's.
Adam(List, float, float, float, float, float, boolean, boolean) - Constructor for class ai.beestream.jitnn.nn.optim.Adam
Adam with the given hyper-parameters; decoupled makes the weight decay AdamW's, amsgrad uses AMSGrad's maximum of the second moment.
AdaMax - Class in ai.beestream.jitnn.nn.optim
AdaMax (Kingma and Ba): Adam with an infinity norm, PyTorch's semantics, L2 weight decay.
AdaMax(List, float, float, float, float, float) - Constructor for class ai.beestream.jitnn.nn.optim.AdaMax
AdaMax with the given betas, eps and L2 weight decay.
adamW(List, float, float) - Static method in class ai.beestream.jitnn.nn.optim.Adam
AdamW: Adam with decoupled weight decay, PyTorch's default betas and eps.
adaptiveAvgPool(Tensor, int...) - Static method in class ai.beestream.jitnn.F
Adaptive average pooling of x [N, C, ...] (1 to 3 spatial axes) to the given output sizes (PyTorch's bins).
AdaptiveAvgPool - Class in ai.beestream.jitnn.nn
Adaptive average pooling to fixed output sizes (1 to 3 spatial axes), PyTorch's nn.AdaptiveAvgPool1d/2d/3d.
AdaptiveAvgPool(int...) - Constructor for class ai.beestream.jitnn.nn.AdaptiveAvgPool
Pools to the output sizes out, one per spatial axis.
adaptiveMaxPool(Tensor, int...) - Static method in class ai.beestream.jitnn.F
Adaptive max pooling of x [N, C, ...] (1 to 3 spatial axes) to the given output sizes.
AdaptiveMaxPool - Class in ai.beestream.jitnn.nn
Adaptive max pooling to fixed output sizes (1 to 3 spatial axes), PyTorch's nn.AdaptiveMaxPool1d/2d/3d.
AdaptiveMaxPool(int...) - Constructor for class ai.beestream.jitnn.nn.AdaptiveMaxPool
Pools to the output sizes out, one per spatial axis.
add(float) - Method in class ai.beestream.jitnn.Tensor
this + c.
add(Tensor) - Method in class ai.beestream.jitnn.Tensor
this + o, element-wise with broadcasting.
add(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
add(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
add(TaskGraph, String) - Method in interface ai.beestream.jitnn.engine.Emitter.TaskAdder
Adds the task to graph as taskId.
Add - Class in ai.beestream.jitnn.nn
Residual connection: relu(main(x) + shortcut(x)) or main(x) + shortcut(x).
Add(Module, Module, boolean) - Constructor for class ai.beestream.jitnn.nn.Add
shortcut null means identity.
addi(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
addi(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
additiveMask() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
An additive attention mask [batch, 1, 1, length]: 0 for tokens, -10000 for padding.
ai.beestream.jitnn - package ai.beestream.jitnn
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).
ai.beestream.jitnn.data - package ai.beestream.jitnn.data
Datasets: MNIST and CIFAR-10 readers, a byte-level text corpus, and synthetic data, held in memory as a Dataset.
ai.beestream.jitnn.dist - package ai.beestream.jitnn.dist
Several devices: collectives (all-reduce, broadcast, reduce-scatter, all-gather) among ranks, staged through the host (HostCollective) or run by NCCL on NVIDIA GPUs (NcclCollective), and ProcessGroup, which runs one rank per device.
ai.beestream.jitnn.djl - package ai.beestream.jitnn.djl
A Deep Java Library engine named jitNN: DJL's NDArrays, blocks, Trainer and Predictor run on jitNN (JitnnEngine).
ai.beestream.jitnn.engine - package ai.beestream.jitnn.engine
The execution engine, which records operators in a per-thread trace and, at a sync point, compiles the trace into one TornadoVM task graph, cached by structure and replayed, or runs it on the host with the double-precision reference.
ai.beestream.jitnn.gguf - package ai.beestream.jitnn.gguf
The GGUF reader (llama.cpp's model format): GgufFile gives a file's metadata and its tensors, memory-mapped without a copy, in their quantised blocks or as F32, F16 and BF16.
ai.beestream.jitnn.io - package ai.beestream.jitnn.io
Checkpoints: named tensors and values (StateDict) saved and loaded in Hugging Face's safetensors format (SafeTensors).
ai.beestream.jitnn.ml - package ai.beestream.jitnn.ml
Classical machine learning after scikit-learn, run on the device: k-means, k-nearest neighbours, PCA, linear and logistic regression, linear SVM, naive Bayes, decision trees, random forests, gradient boosting and t-SNE.
ai.beestream.jitnn.nn - package ai.beestream.jitnn.nn
Neural-network modules after PyTorch's torch.nn: Module and Sequential, linear and convolution layers, pooling, normalisation, attention, embeddings, recurrent layers (LSTM, GRU), graph convolution, activations, and data-parallel training (DataParallel).
ai.beestream.jitnn.nn.models - package ai.beestream.jitnn.nn.models
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.
ai.beestream.jitnn.nn.optim - package ai.beestream.jitnn.nn.optim
Optimisers with PyTorch's semantics (SGD, Adam and AdamW, AdaGrad, RMSProp, AdaDelta, AdaMax, NAdam, LAMB, LARS), learning-rate schedules (LRScheduler) and dynamic loss scaling for FP16 training (GradScaler).
ai.beestream.jitnn.nn.text - package ai.beestream.jitnn.nn.text
Vocabularies and training examples for the embedding models in ai.beestream.jitnn.nn.models.
ai.beestream.jitnn.onnx - package ai.beestream.jitnn.onnx
The ONNX importer: OnnxSession loads an ONNX file and runs it on jitNN's operators.
ai.beestream.jitnn.tf - package ai.beestream.jitnn.tf
The TensorFlow importer: TfModel loads SavedModels and frozen GraphDefs, translating them into ONNX graphs that the ONNX importer runs on jitNN.
alias(Tensor, Shape) - Static method in class ai.beestream.jitnn.Tensor
A tensor that shares base's storage under another shape.
alignment() - Method in class ai.beestream.jitnn.gguf.GgufFile
The data alignment in bytes (general.alignment, 32 by default).
allGather(int, Tensor, Tensor) - Method in interface ai.beestream.jitnn.dist.Collective
input [n, ...] on every rank; output [worldSize * n, ...] receives every rank's input, in rank order.
allGather(int, Tensor, Tensor) - Method in class ai.beestream.jitnn.dist.HostCollective
 
allGather(int, Tensor, Tensor) - Method in class ai.beestream.jitnn.dist.NcclCollective
Persistent output [world * n, ...] receives every rank's input [n, ...], in rank order.
allocateDirect(int) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
allReduce(int, Tensor, Collective.ReduceOp) - Method in interface ai.beestream.jitnn.dist.Collective
Every rank's tensor replaced, in place, by the reduction over all ranks.
allReduce(int, Tensor, Collective.ReduceOp) - Method in class ai.beestream.jitnn.dist.HostCollective
 
allReduce(int, Tensor, Collective.ReduceOp) - Method in class ai.beestream.jitnn.dist.NcclCollective
Persistent tensor replaced, on every rank, by the reduction over the ranks.
allReduced(int, Tensor[], Collective.ReduceOp) - Method in interface ai.beestream.jitnn.dist.Collective
The reduction over all ranks of each tensor.
allReduced(int, Tensor[], Collective.ReduceOp) - Method in class ai.beestream.jitnn.dist.NcclCollective
The reduction over the ranks of each tensor, as new tensors: one NCCL call for all of them.
apply(double) - Method in interface ai.beestream.jitnn.RopeScaling
The frequency used in place of frequency (radians per position).
apply(double) - Method in record class ai.beestream.jitnn.RopeScaling.Linear
 
apply(double) - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
 
apply(float[], int, int[], SplittableRandom) - Method in interface ai.beestream.jitnn.data.Transforms.Transform
 
apply(int[], float[], Tensor...) - Method in class ai.beestream.jitnn.CustomOp
Applies the operator with integer and float attributes (part of the plan's cache key); returns its first output.
apply(Node, Tensor) - Method in interface ai.beestream.jitnn.CustomOp.Backward
The gradient of each input of node (null where none is needed), from the gradient gradOut of its first output.
apply(Tensor...) - Method in class ai.beestream.jitnn.CustomOp
Applies the operator; returns its first output.
apply(Tensor, Tensor, Tensor) - Method in interface ai.beestream.jitnn.CustomOp.UnaryGradient
dL/dx, from the input x, the output y and dL/dy.
apply(Tensor, Tensor, Tensor, Tensor) - Method in interface ai.beestream.jitnn.CustomOp.BinaryGradient
[dL/da, dL/db], from the inputs a and b, the output y and dL/dy.
applyAll(int[], float[], Tensor...) - Method in class ai.beestream.jitnn.CustomOp
Applies the operator; returns all its outputs.
arange(float, float, float) - Static method in class ai.beestream.jitnn.Tensor
start, start + step, ... below end (above end for a negative step): torch.arange, ONNX Range.
arange(float, float, float, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
architecture() - Method in class ai.beestream.jitnn.gguf.GgufFile
general.architecture, for example llama, or null when absent.
argmax(int) - Method in class ai.beestream.jitnn.Tensor
Index of the maximum along axis, as an INT32 tensor (no gradient).
argMax() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
argMax(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
argmin(int) - Method in class ai.beestream.jitnn.Tensor
INT32 index of the first minimum along axis.
argMin() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
argMin(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
argsort(int, boolean) - Method in class ai.beestream.jitnn.Tensor
INT32 indices that sort along axis (stable).
argSort(int, boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
asin() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
asin() - Method in class ai.beestream.jitnn.Tensor
Element-wise arcsine.
asinh() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
asinh() - Method in class ai.beestream.jitnn.Tensor
Element-wise inverse hyperbolic sine.
assign(Tensor) - Method in class ai.beestream.jitnn.Tensor
Records a copy of src into this persistent FP32 tensor of the same size (a running statistic, an optimiser buffer): later operators see the new values.
atan() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
atan() - Method in class ai.beestream.jitnn.Tensor
Element-wise arctangent.
atan2(Tensor) - Method in class ai.beestream.jitnn.Tensor
atan2(this, o), the angle of the point (o, this), with broadcasting.
atan2(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
atanh() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
atanh() - Method in class ai.beestream.jitnn.Tensor
Element-wise inverse hyperbolic tangent.
attention() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns the value of the attention record component.
attentionSoftmax(Tensor, float, boolean) - Static method in class ai.beestream.jitnn.F
softmax(scores * scale) over the last axis of attention scores [..., T, S], in one kernel; causal: query t sees keys 0 .. t, as PyTorch's is_causal.
attr(String) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Attribute key, or null when the node does not set it.
attrFloat(String, float) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Float attribute key, or fallback when the node does not set it.
Attribute(String, int, float, long, String, OnnxModel.TensorData, float[], long[], OnnxModel.Graph) - Constructor for record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Creates an instance of a Attribute record class.
attributes() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns the value of the attributes record component.
attrInt(String, long) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Integer attribute key, or fallback when the node does not set it.
attrInts(String, long[]) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Integer-list attribute key, or fallback when the node does not set it.
attrString(String, String) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
String attribute key, or fallback when the node does not set it.
Autocast - Class in ai.beestream.jitnn
Mixed precision, per thread: inside Autocast.enable(DType), F.linear(Tensor, Tensor, Tensor), Tensor.matmul(Tensor) and the products of F.scaledDotProductAttention(Tensor, Tensor, Tensor, Tensor, boolean, float) cast their operands to FP16 or BF16 and run the half-precision GEMMs, accumulating in FP32.
AUTOTUNE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.autotune.
AUTOTUNE_DIR - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.autotune.dir.
AUTOTUNE_MARGIN - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.autotune.margin.
AUTOTUNE_REPS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.autotune.reps.
AUTOTUNE_RUNS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.autotune.runs.
AUTOTUNE_VERBOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.autotune.verbose.
averageGradients() - Method in class ai.beestream.jitnn.nn.DataParallel
Replaces each parameter's gradient by its mean over the ranks (parameters without a gradient are skipped on every rank alike), with one collective for all of them.
AVG - Enum constant in enum class ai.beestream.jitnn.dist.Collective.ReduceOp
The mean over ranks.
AVG - Enum constant in enum class ai.beestream.jitnn.F.PoolMode
 
avgPool1d(Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
Average pooling of x [N, C, L], PyTorch's avg_pool1d; padding counts in the divisor, as PyTorch's default.
AvgPool1d - Class in ai.beestream.jitnn.nn
Average pooling over 1-D windows, PyTorch's nn.AvgPool1d.
AvgPool1d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.AvgPool1d
Pools windows of kernel with the given stride and padding.
avgPool2d(Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
Average pooling of x [N, C, H, W] over square windows, PyTorch's avg_pool2d; padding counts in the divisor, as PyTorch's default.
AvgPool2d - Class in ai.beestream.jitnn.nn
Average pooling; padding counts in the divisor, as PyTorch's default.
AvgPool2d(int) - Constructor for class ai.beestream.jitnn.nn.AvgPool2d
Pools non-overlapping kernel x kernel windows (stride kernel, no padding).
AvgPool2d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.AvgPool2d
Pools kernel x kernel windows with the given stride and padding.
avgPool3d(Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
Average pooling of x [N, C, D, H, W] over cubic windows, PyTorch's avg_pool3d; padding counts in the divisor.
AvgPool3d - Class in ai.beestream.jitnn.nn
Average pooling over 3-D windows, PyTorch's nn.AvgPool3d.
AvgPool3d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.AvgPool3d
Pools cubic windows of kernel with the given stride and padding.
axis(int) - Method in class ai.beestream.jitnn.Shape
Normalises a possibly negative axis to [0, rank).

B

backend() - Static method in class ai.beestream.jitnn.JitnnRuntime
This thread's backend.
Backend - Enum Class in ai.beestream.jitnn
Where operators run.
BACKEND - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.backend.
backward() - Method in class ai.beestream.jitnn.Tensor
Back-propagates from this scalar.
backward(CustomOp.Backward) - Method in class ai.beestream.jitnn.CustomOp.Builder
The gradient, written with other operators (optional).
backward(Tensor) - Method in class ai.beestream.jitnn.Tensor
Back-propagates from this tensor with gradient (its shape) as the gradient of the result.
BACKWARD - Enum constant in enum class ai.beestream.jitnn.Fft.Norm
The inverse transform scaled by 1/n, the forward unscaled (torch's default).
barrier(int) - Method in interface ai.beestream.jitnn.dist.Collective
Blocks until every rank has called it.
barrier(int) - Method in class ai.beestream.jitnn.dist.HostCollective
 
barrier(int) - Method in class ai.beestream.jitnn.dist.NcclCollective
Runs everything recorded so far, and returns once every rank has.
baseLr - Variable in class ai.beestream.jitnn.nn.optim.LRScheduler
The optimiser's learning rate when the schedule was created.
batch() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns the value of the batch record component.
batch(int, int, float[], int[]) - Method in class ai.beestream.jitnn.data.Blobs
Copies batch b of size samples into xb and yb.
batch(int, int, float[], int[]) - Method in class ai.beestream.jitnn.data.Cifar10
 
batch(int, int, float[], int[]) - Method in interface ai.beestream.jitnn.data.Dataset
Copies batch b of size samples, in dataset order, into x and y.
batch(int, int, float[], int[]) - Method in class ai.beestream.jitnn.data.Mnist
Copies batch b (of size size) of images into x and labels into y.
Batch(int[], int[], int[], int, int) - Constructor for record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Creates an instance of a Batch record class.
Batch(Tensor, Tensor, int, int) - Constructor for record class ai.beestream.jitnn.data.DataLoader.Batch
Creates an instance of a Batch record class.
batchDot(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
batches() - Method in class ai.beestream.jitnn.data.DataLoader
Batches per epoch.
batchMatMul(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
batchNorm(Tensor, Tensor, Tensor, Tensor, Tensor, boolean, float, float) - Static method in class ai.beestream.jitnn.F
Batch normalisation of an NCHW (or NC) tensor.
BatchNorm2d - Class in ai.beestream.jitnn.nn
Batch normalisation over the channels of NCHW (or NC) input, PyTorch defaults: weight 1, bias 0, running mean 0 and variance 1, momentum 0.1, eps 1e-5.
BatchNorm2d(int) - Constructor for class ai.beestream.jitnn.nn.BatchNorm2d
Batch normalisation over channels channels with PyTorch's defaults (momentum 0.1, eps 1e-5).
BatchNorm2d(int, float, float) - Constructor for class ai.beestream.jitnn.nn.BatchNorm2d
Batch normalisation with the given momentum for the running statistics and eps added to the variance.
batchNormRelu(Tensor, Tensor, Tensor, Tensor, Tensor, boolean, float, float) - Static method in class ai.beestream.jitnn.F
relu(batchNorm(...)), fused.
Bert - Class in ai.beestream.jitnn.nn.models
A BERT-style encoder for masked language modelling: token and learned position embeddings with LayerNorm, post-LayerNorm blocks of bidirectional attention (with an additive padding mask) and a GELU MLP of 4 x dim, and an MLM head (dense, GELU, LayerNorm, then the token embedding as the decoder with its own bias).
Bert(int, int, int, int, int, int, int, boolean, float, long) - Constructor for class ai.beestream.jitnn.nn.models.Bert
As above with LayerNorm epsilon eps (RoBERTa uses 1e-5).
Bert(int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.models.Bert
A BERT with an MLP of hidden units, typeVocab token types (0 for none) and, optionally, the pooler (dense and tanh on the first token).
Bert(int, int, int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.models.Bert
An encoder over vocab tokens and up to context positions, with layers blocks of dim features in heads heads, initialised from seed.
bf16(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
bfloat16 tensors are stored as their raw 16-bit patterns.
BF16 - Enum constant in enum class ai.beestream.jitnn.DType
bfloat16: the top 16 bits of an FP32.
BF16 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
bfloat16.
bfloat16() - Method in class ai.beestream.jitnn.Tensor
This tensor converted to BF16 (to(DType.BF16)).
BFLOAT16 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: bfloat16.
bias() - Method in class ai.beestream.jitnn.nn.Conv2d
The bias [out], or null without one.
bias() - Method in class ai.beestream.jitnn.nn.ConvTranspose2d
The bias [out], or null without one.
bias() - Method in class ai.beestream.jitnn.nn.LayerNorm
The shift [features], or null without one.
bias() - Method in class ai.beestream.jitnn.nn.Linear
The bias [out], or null without one.
binary(String, TornadoFunctions.Task5, DoubleBinaryOperator) - Static method in class ai.beestream.jitnn.CustomOp
An element-wise operator on two FP32 tensors of the same shape.
BINARY_REPEAT - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.binaryRepeat.
binaryCrossEntropy(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Binary cross-entropy of probabilities p in [0, 1] against target, PyTorch semantics: the logs are clamped at -100, so p = 0 or 1 gives finite values and gradients.
binaryCrossEntropy(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
F.binaryCrossEntropy(Tensor, Tensor) reduced as reduction says.
binaryCrossEntropyWithLogits(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Binary cross-entropy on logits, sigmoid folded in without overflow (PyTorch's BCEWithLogits).
binaryCrossEntropyWithLogits(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
bitwiseAnd(Tensor) - Method in class ai.beestream.jitnn.Tensor
Bitwise and of INT32 tensors, with broadcasting.
bitwiseNot() - Method in class ai.beestream.jitnn.Tensor
Bitwise complement of an INT32 tensor.
bitwiseOr(Tensor) - Method in class ai.beestream.jitnn.Tensor
Bitwise or of INT32 tensors, with broadcasting.
bitwiseXor(Tensor) - Method in class ai.beestream.jitnn.Tensor
Bitwise exclusive or of INT32 tensors, with broadcasting.
Blobs - Class in ai.beestream.jitnn.data
A synthetic classification set: classes Gaussian clusters in dims dimensions.
Blobs(int, int, int, float, long) - Constructor for class ai.beestream.jitnn.data.Blobs
count samples around classes centres drawn uniformly in [-2, 2)^dims, with Gaussian noise of standard deviation spread, from seed.
blockBytes() - Method in enum class ai.beestream.jitnn.DType
For GGUF block formats: bytes per block.
blockBytes() - Method in enum class ai.beestream.jitnn.gguf.GgmlType
Bytes per block.
blocks(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
GGUF block-quantised tensors: their raw blocks.
blocks1D(int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
groups work-groups of local threads.
blocks2D(int, int, int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
gx * gy work-groups of lx * ly threads.
blocks3D(int, int, int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
gx * gy * gz work-groups of lx threads.
blockValues() - Method in enum class ai.beestream.jitnn.DType
For GGUF block formats: values per block (0 for the other types).
blockValues() - Method in enum class ai.beestream.jitnn.gguf.GgmlType
Values per block (1 for the plain types).
bool(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
A boolean value; throws when the key is absent or holds another type.
BOOL - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: boolean, one byte.
booleanMask(NDArray, int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
broadcast(int, Tensor, int) - Method in interface ai.beestream.jitnn.dist.Collective
Every rank's tensor replaced, in place, by rank root's.
broadcast(int, Tensor, int) - Method in class ai.beestream.jitnn.dist.HostCollective
 
broadcast(int, Tensor, int) - Method in class ai.beestream.jitnn.dist.NcclCollective
Persistent tensor replaced, on every rank, by rank root's.
broadcast(Shape, Shape) - Static method in class ai.beestream.jitnn.Shape
NumPy broadcasting of two shapes.
broadcast(Shape) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
broadcastParameters() - Method in class ai.beestream.jitnn.nn.DataParallel
Copies rank 0's parameters and buffers to every rank.
broadcastStrides(Shape) - Method in class ai.beestream.jitnn.Shape
Strides for reading a tensor of this shape as if it had shape target: broadcast dimensions get stride 0.
Bucketing - Class in ai.beestream.jitnn.nn
Runs models on inputs padded to Buckets sizes, so requests of changing batch and sequence length reuse a few compiled plans, and returns results for the real rows and positions only.
Buckets - Class in ai.beestream.jitnn
A fixed set of sizes that a changing dimension (batch, sequence length) is rounded up to, so requests of many shapes run a few compiled plans instead of compiling one per shape, as TensorRT's optimization profiles or vLLM's CUDA-graph batch sizes do.
buffer(Tensor) - Method in class ai.beestream.jitnn.nn.Module
Registers persistent state that is not trained (running statistics).
buffers() - Method in class ai.beestream.jitnn.nn.Module
This module's buffers and its children's.
build() - Method in class ai.beestream.jitnn.CustomOp.Builder
The operator, registered under its name.
build() - Method in class ai.beestream.jitnn.data.DataLoader.Builder
The loader with these options.
build() - Method in class ai.beestream.jitnn.JitnnConfig.Builder
The configuration.
build(List, int) - Static method in class ai.beestream.jitnn.nn.text.Vocabulary
The vocabulary of the words in sentences seen at least minCount times.
builder() - Static method in class ai.beestream.jitnn.JitnnConfig
A builder starting from the defaults.
builder(Dataset, int) - Static method in class ai.beestream.jitnn.data.DataLoader
A builder for a loader of data in batches of batchSize.
builder(String) - Static method in class ai.beestream.jitnn.CustomOp
Starts the definition of an operator named name.
builderFromSystemProperties() - Static method in class ai.beestream.jitnn.JitnnConfig
A builder starting from the system properties.
bytes() - Method in enum class ai.beestream.jitnn.DType
Bytes per element: 4, 2, or 1 for the 8-bit and block formats (whose blocks DType.blockBytes() sizes).
bytes() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns the value of the bytes record component.
bytes() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns the value of the bytes record component.
bytes() - Method in class ai.beestream.jitnn.onnx.Proto
The current length-delimited field as a view (no copy).
byteSize(long) - Method in enum class ai.beestream.jitnn.gguf.GgmlType
Bytes taken by elements values (elements a multiple of the block size).

C

c() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
Returns the value of the c record component.
cat(int, Tensor...) - Static method in class ai.beestream.jitnn.Tensor
Joins tensors along an existing axis.
causal(Buckets, Buckets, Tensor, int, UnaryOperator) - Static method in class ai.beestream.jitnn.nn.Bucketing
A causal token model (GPT-2, Llama) over INT32 tokens [B, T]: rows padded to a batch size, positions padded at the end to a seq size with padToken, which the real positions never attend to; the result [B, T, ...] for the real rows and positions.
CBOW - Enum constant in enum class ai.beestream.jitnn.nn.models.Word2Vec.Mode
The mean of the context predicts the centre word.
cbowSamples(int) - Method in class ai.beestream.jitnn.nn.text.Corpus
CBOW samples, only positions with a full window on both sides, shuffled: [sentence, centre, context (2 window per sample, flattened)].
cbrt() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
ceil() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
ceil() - Method in class ai.beestream.jitnn.Tensor
Element-wise ceiling.
ceilDiv(int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
a / b rounded up, for grid sizes.
celu(float) - Method in class ai.beestream.jitnn.Tensor
CELU: max(0, x) + min(0, alpha (exp(x / alpha) - 1)).
CELU - Class in ai.beestream.jitnn.nn
CELU with alpha (PyTorch's default 1).
CELU(float) - Constructor for class ai.beestream.jitnn.nn.CELU
CELU with the given alpha.
centerCrop(Tensor, int, int) - Static method in class ai.beestream.jitnn.F
The central height x width window of images [..., H, W] (torchvision's center_crop, top-left rounding down).
centres() - Method in class ai.beestream.jitnn.ml.KMeans
The cluster centres [k, features].
change(JitnnConfig.Option, T) - Static method in class ai.beestream.jitnn.JitnnConfig
Changes an option that may change at run time (JitnnConfig.Option.changeableAtRunTime()); operations recorded afterwards see the new value.
changeableAtRunTime() - Method in class ai.beestream.jitnn.JitnnConfig.Option
Whether the option may change after the runtime has started (JitnnConfig.change(JitnnConfig.Option, T)).
CHANNELS_LAST - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.channelsLast.
CHANNELS_LAST_VERBOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.channelsLast.verbose.
CharRnn - Class in ai.beestream.jitnn.nn.models
A recurrent language model: token embedding, a stack of LSTM or GRU layers, a linear head over the vocabulary.
CharRnn(int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.models.CharRnn
A model over vocab tokens: embeddings of dim, layers LSTM (or GRU, with useGru) layers of hidden features, and the head.
check(String, Tensor) - Method in class ai.beestream.jitnn.io.StateDict
Throws unless tensor name can be copied into target (see StateDict.copyInto(String, Tensor)).
checkpoint(Function, Tensor...) - Static method in class ai.beestream.jitnn.Grad
Activation checkpointing, as PyTorch's torch.utils.checkpoint: runs segment on inputs and keeps none of its activations for the backward pass, which runs the segment again to get them.
checkpointing(boolean) - Method in class ai.beestream.jitnn.nn.models.Bert
With on, training keeps no activations inside the blocks: each block runs again in the backward pass (Grad.checkpoint(Function, Tensor...)), for less memory and about a third more compute.
checkpointing(boolean) - Method in class ai.beestream.jitnn.nn.models.GPT2
With on, training keeps no activations inside the blocks: each block runs again in the backward pass (Grad.checkpoint(Function, Tensor...)), for less memory and about a third more compute.
checkpointing(boolean) - Method in class ai.beestream.jitnn.nn.models.Llama
With on, training keeps no activations inside the blocks: each block runs again in the backward pass (Grad.checkpoint(Function, Tensor...)), for less memory and about a third more compute.
child(M) - Method in class ai.beestream.jitnn.nn.Module
Adds a child module; returns it.
choices() - Method in class ai.beestream.jitnn.JitnnConfig.Option
The accepted values of an option with a fixed set of them, else an empty list.
cholesky(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Lower Cholesky factor L, A = L L^T, of symmetric positive-definite matrices (only the lower triangle is read).
chunk(int, int) - Method in class ai.beestream.jitnn.Tensor
Splits axis into chunks equal pieces.
Cifar10 - Class in ai.beestream.jitnn.data
CIFAR-10 from the binary distribution (cifar-10-batches-bin: data_batch_1..5.bin and test_batch.bin, each record a label byte and 3072 pixel bytes in CHW order).
clamp(float, float) - Method in class ai.beestream.jitnn.Tensor
x clamped to [lo, hi].
CLASSIFICATION - Enum constant in enum class ai.beestream.jitnn.ml.RandomForest.Task
Class labels, by averaged class probabilities.
clip(Number, Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
clipGradNorm(float) - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Clips the gradients of this optimiser's parameters to a global L2 norm of at most maxNorm at the next Optimizer.step(), as PyTorch's clip_grad_norm_, and returns the norm before clipping (after loss unscaling).
clipNorm(List, float) - Static method in class ai.beestream.jitnn.Grad
Rescales the gradients of params so that their global L2 norm is at most maxNorm, as PyTorch's clip_grad_norm_: every gradient is multiplied by min(1, maxNorm / (norm + 1e-6)).
close() - Method in class ai.beestream.jitnn.data.DataLoader
 
close() - Method in class ai.beestream.jitnn.dist.NcclCollective
 
close() - Method in class ai.beestream.jitnn.djl.JitnnModel
Releases the model's weights before closing it, so that loading the same model again reuses the plans compiled for this one.
close() - Method in interface ai.beestream.jitnn.Grad.Scope
 
close() - Method in class ai.beestream.jitnn.nn.Module
Releases the parameters and buffers of this module and its children, their gradients, and the weights jitNN derived from them (BatchNorm-folded and channels-last copies).
close() - Method in class ai.beestream.jitnn.onnx.OnnxSession
Releases the model's weights: plans compiled for this session can then run another session of the same model (loading it again) without compiling.
close() - Method in class ai.beestream.jitnn.Tensor
Tensor.release(), for try-with-resources.
close() - Method in class ai.beestream.jitnn.tf.TfModel
Releases the model's weights (see OnnxSession.close()); the model cannot run afterwards.
coef() - Method in class ai.beestream.jitnn.ml.LinearRegression
Coefficients [d] (one target) or [t, d].
coef() - Method in class ai.beestream.jitnn.ml.LinearSvc
The weights: [1, features] for two classes, else [classes, features].
coef() - Method in class ai.beestream.jitnn.ml.LogisticRegression
Coefficients [1, d] for two classes, [c, d] otherwise.
Collective - Interface in ai.beestream.jitnn.dist
Collective communication among the ranks of a group, each rank one thread and usually one device (after torch.distributed).
Collective.ReduceOp - Enum Class in ai.beestream.jitnn.dist
How a reduction combines the ranks' tensors, element-wise.
cols() - Method in class ai.beestream.jitnn.SparseMatrix
The number of columns.
components() - Method in class ai.beestream.jitnn.ml.Pca
The components [k, d].
CONCURRENCY - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.concurrency.
config() - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
The parsed config.json.
configFloat(String, float) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Config value key as a float, or otherwise when absent.
configInt(String...) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
The first of keys present in the config, as an int; throws when none is.
configString(String, String) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Config value key as a string, or otherwise when absent or null.
constant(int...) - Method in class ai.beestream.jitnn.engine.Emitter
A constant int array uploaded once with the graph (shape and stride metadata).
CONSTANT - Enum constant in enum class ai.beestream.jitnn.F.PadMode
Fills with a constant value.
constantF32(float...) - Method in class ai.beestream.jitnn.engine.Emitter
A constant float array uploaded once with the graph.
contentEquals(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
context() - Method in class ai.beestream.jitnn.engine.Emitter
A new KernelContext, the first argument of every kernel.
contrastiveLoss(Tensor, Tensor, Tensor, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Contrastive loss (Hadsell et al., 2006) with similar 1 for similar pairs and 0 for dissimilar ones: similar d^2 + (1 - similar) max(margin - d, 0)^2.
conv(Tensor, Tensor, Tensor, int[], int[], int[], int) - Static method in class ai.beestream.jitnn.F
Convolution over 1 to 3 spatial axes with a size per axis: pads holds the padding before every axis, then after every axis (ONNX's order), so it may be asymmetric.
CONV - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.conv.
CONV_PHASES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.convPhases.
CONV_WEIGHT_CHUNK - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.conv.weightChunk.
CONV_WEIGHT_GROUPS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.conv.weightGroups.
conv1d(Tensor, Tensor, Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
1-D convolution of x [N, C, L] with weight [Cout, C, K]; bias may be null.
conv1d(Tensor, Tensor, Tensor, int, int, int, int) - Static method in class ai.beestream.jitnn.F
Grouped 1-D convolution: weight [Cout, C / groups, K]; groups = C is depthwise.
Conv1d - Class in ai.beestream.jitnn.nn
1-D convolution over [N, C, ...] with a cubic kernel, optionally grouped, PyTorch's initialisation: weight and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = in * kernel^1.
Conv1d(int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv1d
A 1-D convolution from in to out channels, initialised from seed.
Conv1d(int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv1d
With groups groups of in / groups input channels each (fan_in = in / groups * kernel^1).
CONV1X1 - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.conv1x1.
CONV1X1_BACKWARD - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.conv1x1.backward.
CONV1X1_GEMM - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.conv1x1.gemm.
conv2d(Tensor, Tensor, Tensor, int, int, int, int) - Static method in class ai.beestream.jitnn.F
2-D convolution (NCHW), square kernel; bias may be null.
Conv2d - Class in ai.beestream.jitnn.nn
2-D convolution over NCHW with a square kernel, PyTorch's nn.Conv2d semantics and initialisation: weight and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = in/groups * k * k.
Conv2d(int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv2d
A convolution from in to out channels in groups groups (both divisible by it), initialised from seed.
Conv2d(int, int, int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.Conv2d
A convolution from in to out channels with a kernel x kernel kernel and a bias, no dilation, one group.
conv2dRelu(Tensor, Tensor, Tensor, int, int, int, int) - Static method in class ai.beestream.jitnn.F
relu(conv2d(...)) with the bias and ReLU fused into the convolution's output.
conv3d(Tensor, Tensor, Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
3-D convolution of x [N, C, D, H, W] with weight [Cout, C, KD, KH, KW]; bias may be null.
conv3d(Tensor, Tensor, Tensor, int, int, int, int) - Static method in class ai.beestream.jitnn.F
Grouped 3-D convolution: weight [Cout, C / groups, KD, KH, KW].
Conv3d - Class in ai.beestream.jitnn.nn
3-D convolution over [N, C, ...] with a cubic kernel, optionally grouped, PyTorch's initialisation: weight and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = in * kernel^3.
Conv3d(int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv3d
A 3-D convolution from in to out channels with a cubic kernel, initialised from seed.
Conv3d(int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv3d
With groups groups of in / groups input channels each (fan_in = in / groups * kernel^3).
convTranspose2d(Tensor, Tensor, Tensor, int, int, int, int, int) - Static method in class ai.beestream.jitnn.F
Transposed 2-D convolution (NCHW), PyTorch's conv_transpose2d: weight [Cin, Cout/groups, K, K], square kernel; bias may be null.
ConvTranspose2d - Class in ai.beestream.jitnn.nn
Transposed 2-D convolution over NCHW with a square kernel, PyTorch's nn.ConvTranspose2d semantics and initialisation: weight [in, out/groups, k, k] and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = out/groups * k * k.
ConvTranspose2d(int, int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.ConvTranspose2d
A transposed convolution from in to out channels in groups groups, initialised from seed.
ConvTranspose2d(int, int, int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.ConvTranspose2d
A transposed convolution from in to out channels with a bias, no output padding or dilation, one group.
cooccurrence(int) - Method in class ai.beestream.jitnn.nn.text.Corpus
Symmetric co-occurrence counts within window, each weighted 1 / distance: [rows, cols, value].
copyFrom(float[]) - Method in class ai.beestream.jitnn.Tensor
Overwrites a persistent tensor's contents from the host (e.g. loading weights).
copyInto(String, Tensor) - Method in class ai.beestream.jitnn.io.StateDict
Overwrites the persistent tensor target (a parameter, a buffer, optimiser state) with tensor name, in place, so plans compiled with target keep using it.
Corpus - Class in ai.beestream.jitnn.nn.text
Training examples for the embedding models, drawn on the host from sentences of word ids: skip-gram pairs, CBOW windows, negative samples from the unigram distribution raised to 3/4, and GloVe's distance-weighted co-occurrence counts.
Corpus(List, long[], long) - Constructor for class ai.beestream.jitnn.nn.text.Corpus
Examples from sentences of word ids, negatives drawn by the word counts counts (indexed by id), randomness from seed.
cos() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
cos() - Method in class ai.beestream.jitnn.Tensor
Element-wise cosine.
cosh() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
cosh() - Method in class ai.beestream.jitnn.Tensor
Element-wise hyperbolic cosine.
cosine(float[], int, int, int) - Static method in class ai.beestream.jitnn.nn.text.Corpus
Cosine similarity of rows a and b of vectors [n, dim].
cosine(Optimizer, int, float) - Static method in class ai.beestream.jitnn.nn.optim.LRScheduler
Cosine annealing from base to minLr over tMax epochs (PyTorch's closed form).
cosineEmbeddingLoss(Tensor, Tensor, Tensor, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Cosine embedding loss: 1 - cos(a, b) for target 1, max(0, cos(a, b) - margin) for -1.
cosineProximityLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Cosine proximity loss (DL4J, Keras): minus the cosine similarity along the last axis, reduced.
cosineSimilarity(Tensor, Tensor, int, float) - Static method in class ai.beestream.jitnn.F
Cosine similarity along axis: a . b / (max(|a|, eps) max(|b|, eps)), PyTorch's cosine_similarity.
count - Variable in class ai.beestream.jitnn.data.Blobs
The number of samples.
count - Variable in class ai.beestream.jitnn.data.Cifar10
The number of images.
count - Variable in class ai.beestream.jitnn.data.Mnist
The number of images.
count() - Method in class ai.beestream.jitnn.data.Cifar10
 
count() - Method in interface ai.beestream.jitnn.data.Dataset
The number of samples.
count() - Method in class ai.beestream.jitnn.data.Mnist
 
counts() - Method in class ai.beestream.jitnn.nn.text.Vocabulary
Each word's count, indexed by id.
CPU - Enum constant in enum class ai.beestream.jitnn.Backend
Operators are recorded lazily as on Backend.TORNADO; a sync point rewrites the trace with the same fusion pass and runs it on the host, each operator with its multithreaded, vectorised CPU kernel (Java Vector API) or, when it has none, its reference implementation.
CPU_BLAS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.blas.
CPU_BNNS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.bnns.
CPU_GEMM_KC - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.gemm.kc.
CPU_GEMM_MC_PANELS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.gemm.mcPanels.
CPU_GEMM_NC - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.gemm.nc.
CPU_GEMM_TILE_MACS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.gemm.tileMacs.
CPU_NATIVE_CHOOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.native.choose.
CPU_ONEDNN - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.onednn.
CPU_OPENCL - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.opencl.
CPU_OPENCL_CHOOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.opencl.choose.
CPU_POOL_BYTES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.poolBytes.
CPU_REUSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.reuse.
CPU_REUSE_TRACES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.reuse.traces.
CPU_SPIN_MICROS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.spinMicros.
CPU_STATS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.stats.
CPU_STATS_SHAPES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.stats.shapes.
CPU_THREADS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cpu.threads.
cpuOpenClDevice() - Static method in class ai.beestream.jitnn.JitnnRuntime
The OpenCL device on the host CPU (<driver>:<device>) where the CPU backend runs the operators it has no kernel for, or null when there is none.
cpuReport() - Static method in class ai.beestream.jitnn.JitnnRuntime
The CPU backend's operators by time since the last call (as printed at exit with jitnn.cpu.stats), then starts counting again; empty when nothing ran.
cpuStats() - Static method in class ai.beestream.jitnn.JitnnRuntime
The CPU backend's counters so far, per operator name: [calls, nanoseconds, calls that ran the reference implementation, calls that ran on the host's OpenCL device]; the times only with jitnn.cpu.stats=true.
create(Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
create(Buffer, Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
crossEntropy(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Mean cross-entropy between logits [N, C] and INT32 class labels [N] (fused kernels).
crossEntropy(Tensor, Tensor) - Static method in class ai.beestream.jitnn.nn.Bucketing
Mean cross-entropy of the first N rows of logits [B, C] (B a padded batch) against INT32 labels [N]: each real row weighs 1/N and each padded row 0, through a weight tensor, so every N in the bucket runs the same plan.
ctcLoss(Tensor, Tensor, int[], int[], int, F.Reduction) - Static method in class ai.beestream.jitnn.F
CTC loss, PyTorch's ctc_loss: log-probabilities [T, B, C] (log_softmax over C), INT32 targets [B, L] padded, input and target lengths; blank index blank.
cube() - Method in class ai.beestream.jitnn.Tensor
x^3 (DL4J's Cube activation).
Cube - Class in ai.beestream.jitnn.nn
x^3.
Cube() - Constructor for class ai.beestream.jitnn.nn.Cube
 
CUDA_GRAPH - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.cudaGraph.
cumprod(int) - Method in class ai.beestream.jitnn.Tensor
Inclusive cumulative product along axis; its gradient is exact with zeros.
cumProd(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
cumProd(int, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
cumsum(int) - Method in class ai.beestream.jitnn.Tensor
Inclusive cumulative sum along axis.
cumSum() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
cumSum(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
current() - Static method in class ai.beestream.jitnn.Autocast
The active autocast type, or null.
current() - Static method in class ai.beestream.jitnn.JitnnConfig
The runtime's configuration.
currentRate() - Method in class ai.beestream.jitnn.nn.optim.LRScheduler
The optimiser's current learning rate.
CustomOp - Class in ai.beestream.jitnn
An operator defined outside jitNN: a kernel written in plain Java with TornadoVM's KernelContext, an optional host reference and an optional gradient.
CustomOp.Backward - Interface in ai.beestream.jitnn
Gradients with respect to each input (null where none is needed), given the gradient of the first output.
CustomOp.BinaryGradient - Interface in ai.beestream.jitnn
The gradients of an element-wise binary operator y = f(a, b): [dL/da, dL/db] from a, b, y and dL/dy.
CustomOp.Builder - Class in ai.beestream.jitnn
Defines a custom operator step by step; CustomOp.Builder.build() registers its name.
CustomOp.Launcher - Interface in ai.beestream.jitnn
Adds the operator's device tasks for node to the plan being built.
CustomOp.OutputShape - Interface in ai.beestream.jitnn
The shape of an output, from the inputs and the integer attributes of the call.
CustomOp.Reference - Interface in ai.beestream.jitnn
Computes node on the host: inputs are readable, outputs are allocated.
CustomOp.UnaryGradient - Interface in ai.beestream.jitnn
The gradient of an element-wise unary operator y = f(x): dL/dx from x, y and dL/dy.
cut(Tensor, int, int) - Static method in class ai.beestream.jitnn.Buckets
The first length entries of dimension axis of t, copied on the host (a device result is read back first).

D

data() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
Returns the value of the data record component.
DataLoader - Class in ai.beestream.jitnn.data
Batches of a Dataset for training, as PyTorch's DataLoader: shuffled per epoch from a seed, optional per-sample transforms (augmentation), the last short batch kept or dropped, and batches assembled ahead by worker threads while the device computes.
DataLoader.Batch - Record Class in ai.beestream.jitnn.data
One batch: samples [size, ...sampleShape] and labels [size], host tensors.
DataLoader.Builder - Class in ai.beestream.jitnn.data
Options of a DataLoader: shuffling, worker threads, prefetching.
dataOffset() - Method in class ai.beestream.jitnn.gguf.GgufFile
Byte offset of the tensor data section.
DataParallel - Class in ai.beestream.jitnn.nn
Data-parallel training with one model replica per rank (see ProcessGroup): broadcast rank 0's parameters once, then after each backward pass average the gradients over the ranks before the optimiser step, so every replica takes the same step, as PyTorch's DistributedDataParallel does.
DataParallel(Module, int, Collective) - Constructor for class ai.beestream.jitnn.nn.DataParallel
Wraps rank rank's replica model; ranks synchronise through collective.
Dataset - Interface in ai.beestream.jitnn.data
An in-memory labelled dataset of fixed-shape samples (CHW images or flat vectors).
dataType() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Returns the value of the dataType record component.
Dcgan - Class in ai.beestream.jitnn.nn.models
DCGAN (Radford et al., 2016) for 28x28 single-channel images: a generator from a latent vector through a 7x7 feature map and two stride-2 transposed convolutions to a tanh image, and a discriminator of two stride-2 convolutions with LeakyReLU(0.2) and a linear logit.
Dcgan.Discriminator - Class in ai.beestream.jitnn.nn.models
Images [B, 1, 28, 28] to logits [B, 1].
Dcgan.Generator - Class in ai.beestream.jitnn.nn.models
z [B, latent] to images [B, 1, 28, 28] in [-1, 1].
DEBUG_KEYS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.debugKeys.
decisionFunction(Tensor) - Method in class ai.beestream.jitnn.ml.GradientBoosting
Raw scores [m, outputs]: the baseline plus every tree.
decisionFunction(Tensor) - Method in class ai.beestream.jitnn.ml.LinearSvc
The decision values x w^T + b of each sample.
decisionFunction(Tensor) - Method in class ai.beestream.jitnn.ml.LogisticRegression
The decision values x W^T + b of each sample.
decode(int[], boolean) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
The text of ids; with skipSpecial, special added tokens are left out.
decode(Tensor) - Method in class ai.beestream.jitnn.nn.models.Vae
Decodes latent vectors z [B, latent] to [B, in], each value in (0, 1).
defaultDevice() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
defaultIfUnset(JitnnConfig.Option, T) - Static method in class ai.beestream.jitnn.JitnnConfig
Gives option a different default before the runtime starts, unless it was set: for integrations such as the DJL engine.
defaults() - Static method in class ai.beestream.jitnn.JitnnConfig
All defaults.
defaultText() - Method in class ai.beestream.jitnn.JitnnConfig.Option
The default as the documentation states it (Markdown).
defaultValue() - Method in class ai.beestream.jitnn.JitnnConfig.Option
The default, or null when it depends on the machine or on other options.
depthToSpace(Tensor, int, String) - Static method in class ai.beestream.jitnn.F
Moves blocks of channels into space: x [N, C, H, W] to [N, C / b^2, H b, W b].
dequantize(Tensor) - Static method in class ai.beestream.jitnn.F
A Q8_0 tensor back to FP32.
describeInput() - Method in class ai.beestream.jitnn.djl.JitnnSymbolBlock
 
description() - Method in class ai.beestream.jitnn.JitnnConfig.Option
What the option does (Markdown).
det(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Determinants [...] of square matrices.
detach() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
detach() - Method in class ai.beestream.jitnn.Tensor
The same data without gradient history.
deterministic() - Static method in class ai.beestream.jitnn.JitnnRuntime
Whether deterministic results were asked for (JitnnRuntime.setDeterministic(boolean), jitnn.deterministic).
DETERMINISTIC - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.deterministic.
DEVICE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.device.
dim() - Method in class ai.beestream.jitnn.nn.models.Doc2Vec
The vector size.
dim() - Method in class ai.beestream.jitnn.nn.models.GloVe
The vector size.
dim() - Method in class ai.beestream.jitnn.nn.models.Word2Vec
The vector size.
dim(int) - Method in class ai.beestream.jitnn.Shape
Dimension axis; negative values count from the end.
dim(int) - Method in class ai.beestream.jitnn.Tensor
The size of axis axis; a negative axis counts from the end.
dimNames() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Returns the value of the dimNames record component.
dims - Variable in class ai.beestream.jitnn.data.Blobs
The number of dimensions.
dims() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Returns the value of the dims record component.
dims() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Returns the value of the dims record component.
dims() - Method in class ai.beestream.jitnn.Shape
A copy of the sizes.
directory() - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
The directory the checkpoint was read from.
disabled() - Static method in class ai.beestream.jitnn.Grad
Disables gradient recording until the returned scope is closed.
Discriminator(int, long) - Constructor for class ai.beestream.jitnn.nn.models.Dcgan.Discriminator
A discriminator with width channels in its first layer, initialised from seed.
div(Tensor) - Method in class ai.beestream.jitnn.Tensor
this / o, element-wise with broadcasting.
div(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
div(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
divi(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
divi(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
DJL_FALLBACK - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.djl.fallback.
DJL_PERSISTENT_BYTES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.djl.persistentBytes.
Doc2Vec - Class in ai.beestream.jitnn.nn.models
Paragraph vectors (Le and Mikolov, 2014) with negative sampling.
Doc2Vec(int, int, int, Doc2Vec.Mode, long) - Constructor for class ai.beestream.jitnn.nn.models.Doc2Vec
Paragraph vectors for documents documents over vocab words, vectors of dim, trained as mode says, initialised from seed.
Doc2Vec.Mode - Enum Class in ai.beestream.jitnn.nn.models
How a document's words are predicted.
documentVectors() - Method in class ai.beestream.jitnn.nn.models.Doc2Vec
All document vectors [documents, dim] on the host.
domain() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns the value of the domain record component.
dot(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
DOUBLE - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 64-bit float.
doubles(Proto.DoubleSink) - Method in class ai.beestream.jitnn.onnx.Proto
Repeated double values: one fixed64, or a packed run.
dropLast(boolean) - Method in class ai.beestream.jitnn.data.DataLoader.Builder
Leaves out the last batch when it is short.
dropout(Tensor, float, Tensor, int) - Static method in class ai.beestream.jitnn.F
Inverted dropout with a fresh mask per step: seed is an INT32 [2] host tensor.
Dropout - Class in ai.beestream.jitnn.nn
Inverted dropout in training mode, identity in evaluation mode.
Dropout(float, long) - Constructor for class ai.beestream.jitnn.nn.Dropout
Dropout that zeroes each element with probability p and scales the rest by 1 / (1 - p); masks are seeded from seed.
dtype() - Method in enum class ai.beestream.jitnn.gguf.GgmlType
The jitNN type this one loads as, or null when jitNN has none.
dtype() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns the value of the dtype record component.
dtype() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
Returns the value of the dtype record component.
dtype() - Method in class ai.beestream.jitnn.Tensor
The element type of this tensor.
dtype(String) - Method in class ai.beestream.jitnn.io.StateDict
The dtype of tensor name.
DType - Enum Class in ai.beestream.jitnn
Element type of a tensor: FP32 data, INT32 labels and indices, and the half-precision storage types for mixed precision, IEEE FP16 and bfloat16 (the top 16 bits of an FP32), and Q8_0, GGUF's 8-bit blocks (per 32 consecutive values an FP16 scale and 32 signed bytes, 34 bytes), and the OCP 8-bit floats FP8 E4M3 and E5M2, one byte per value, stored without their scale (see Fp8).
dummyF32() - Method in class ai.beestream.jitnn.engine.Emitter
A one-element array for optional array arguments that a kernel does not read.
duplicate() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
DYNAMIC_SHAPES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.dynamicShapes.

E

eigh(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Eigenvalues (ascending) [..., n] and eigenvectors (columns) [..., n, n] of symmetric matrices, torch.linalg.eigh.
eigvalsh(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Eigenvalues (ascending) of symmetric matrices, torch.linalg.eigvalsh.
elemType() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Returns the value of the elemType record component.
elu(float) - Method in class ai.beestream.jitnn.Tensor
ELU: x for x > 0, else alpha (exp(x) - 1).
ELU - Class in ai.beestream.jitnn.nn
ELU with slope alpha for negative inputs (PyTorch's default 1).
ELU(float) - Constructor for class ai.beestream.jitnn.nn.ELU
ELU with the given alpha.
embedding(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Rows of weight [V, D] picked by the INT32 index: shape index.shape + [D].
embedding(Tensor, Tensor, int) - Static method in class ai.beestream.jitnn.F
As F.embedding(Tensor, Tensor); row paddingIdx receives no gradient.
Embedding - Class in ai.beestream.jitnn.nn
A lookup table of count vectors of size dim, applied to INT32 indices.
Embedding(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.Embedding
Row paddingIdx receives no gradient (pass -1 for none).
Embedding(int, int, long) - Constructor for class ai.beestream.jitnn.nn.Embedding
A table of count vectors of size dim, standard normal from seed, every row trained.
Embedding(Tensor, int) - Constructor for class ai.beestream.jitnn.nn.Embedding
An embedding with the given initial table [count, dim].
embeddingBag(Tensor, int[], Tensor, F.BagMode) - Static method in class ai.beestream.jitnn.F
EmbeddingBag (PyTorch's, 1-D input with offsets): the rows of weight picked by the INT32 indices, combined per bag by mode.
EmbeddingBag - Class in ai.beestream.jitnn.nn
EmbeddingBag: bags of INT32 indices (one flat list plus bag offsets) to the sum, mean or max of their embedding rows, PyTorch's nn.EmbeddingBag with standard normal initial weights.
EmbeddingBag(int, int, F.BagMode, long) - Constructor for class ai.beestream.jitnn.nn.EmbeddingBag
A table of count vectors of size dim, standard normal from seed, whose bags combine as mode says.
emit(Emitter, Node) - Method in interface ai.beestream.jitnn.CustomOp.Launcher
Adds the tasks that compute node through e.
Emitter - Class in ai.beestream.jitnn.engine
What an operator sees while its device tasks are added to a task graph: the arrays bound to its tensors, constant metadata arrays, and helpers for launch grids.
Emitter.TaskAdder - Interface in ai.beestream.jitnn.engine
Adds one task to the graph under the given task id.
enable(DType) - Static method in class ai.beestream.jitnn.Autocast
Enables autocast to dtype (F16 or BF16) until the returned scope is closed.
enableHalf() - Static method in class ai.beestream.jitnn.Autocast
Half-precision activations, as a model converted to FP16: inside the scope, linear layers, element-wise operators, LayerNorm and RMSNorm and attention take and return FP16 tensors (computing in FP32 inside), so activations stay FP16 between them, with half the memory traffic.
encode(Tensor, Tensor) - Method in class ai.beestream.jitnn.nn.models.Bert
Hidden states [B, T, dim]; mask is additive, broadcast to [B, heads, T, T] (e.g.
encode(Tensor, Tensor, Tensor) - Method in class ai.beestream.jitnn.nn.models.Bert
Hidden states [B, T, dim] with INT32 token types [B, T] (null for all 0; ignored by a model without token-type embeddings) and an additive mask (null for none).
encode(String) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
Encodes one text, with the post-processor's special tokens.
encode(String, String) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
Encodes a pair of texts (sentence A, sentence B).
encode(String, String, int) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
Encodes a (and b when not null) into at most maxLength ids, special tokens included, cutting the longer sequence first as Hugging Face's longest-first truncation.
encode(List) - Method in class ai.beestream.jitnn.nn.text.Vocabulary
Each sentence as ids, words outside the vocabulary dropped.
encodeBatch(List, int) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
Encodes texts, each cut to maxLength, padded to the longest with Tokenizer.padId().
Encoding(int[], int[], String[]) - Constructor for record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Creates an instance of a Encoding record class.
ENGINE_NAME - Static variable in class ai.beestream.jitnn.djl.JitnnEngine
The engine's name in DJL: Engine.getEngine("jitNN").
entries() - Method in record class ai.beestream.jitnn.io.SafeTensors.Header
Returns the value of the entries record component.
Entry(DType, String, long[], long, long) - Constructor for record class ai.beestream.jitnn.io.SafeTensors.Entry
Creates an instance of a Entry record class.
epoch - Variable in class ai.beestream.jitnn.nn.optim.LRScheduler
Epochs (or, for warm-up schedules, steps) counted so far.
epoch(int) - Method in class ai.beestream.jitnn.data.DataLoader
The batches of epoch epoch (its shuffle and augmentation draws).
eq(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where this == o, else 0, with broadcasting.
eq(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
eq(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
equals(Object) - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.io.SafeTensors.Header
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in class ai.beestream.jitnn.JitnnConfig
 
equals(Object) - Method in record class ai.beestream.jitnn.nn.GRU.Output
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.nn.LSTM.Output
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.nn.models.Vae.Output
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.RopeScaling.Linear
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Indicates whether some other object is "equal to" this one.
equals(Object) - Method in class ai.beestream.jitnn.Shape
 
equals(Object) - Method in record class ai.beestream.jitnn.SparseMatrix.Valued
Indicates whether some other object is "equal to" this one.
erf() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
erf() - Method in class ai.beestream.jitnn.Tensor
Element-wise error function.
erfc() - Method in class ai.beestream.jitnn.Tensor
Element-wise complementary error function.
erfinv() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
evalWindow(int, int, int[], int[], int) - Method in class ai.beestream.jitnn.data.TextCorpus
Evaluation window w: inputs and targets [seq].
evalWindows(int) - Method in class ai.beestream.jitnn.data.TextCorpus
Number of non-overlapping evaluation windows of seq tokens.
exp() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
exp() - Method in class ai.beestream.jitnn.Tensor
e^x, element-wise.
expand(Shape) - Method in class ai.beestream.jitnn.Tensor
Broadcast copy to target shape.
expandDims(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
explainedVariance() - Method in class ai.beestream.jitnn.ml.Pca
The variance explained by each component.
explainedVarianceRatio() - Method in class ai.beestream.jitnn.ml.Pca
The fraction of the total variance explained by each component.
expm1() - Method in class ai.beestream.jitnn.Tensor
Element-wise exp(x) - 1, accurate near 0.
export(Path, Function, Tensor) - Static method in class ai.beestream.jitnn.onnx.OnnxExport
export(Path, Function, Tensor[], String[], String) - Static method in class ai.beestream.jitnn.onnx.OnnxExport
Records forward on inputs (put the model in evaluation mode first) and writes it to file with graph inputs inputNames and the output outputName.
eye(int, int, int) - Static method in class ai.beestream.jitnn.Tensor
The n x m matrix with ones on diagonal k (0 main, positive above, negative below), zeros elsewhere.
eye(int, int, int, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
eyeLike(Tensor, int) - Static method in class ai.beestream.jitnn.Tensor
Tensor.eye(int, int, int) with the shape of the rank-2 tensor like (ONNX EyeLike).

F

f() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the f record component.
F - Class in ai.beestream.jitnn
Functional operators that are not methods of Tensor.
F.BagMode - Enum Class in ai.beestream.jitnn
F.PadMode - Enum Class in ai.beestream.jitnn
How F.pad(Tensor, int[], int[], F.PadMode, float) fills: a constant, the reflection without the edge, or the edge replicated.
F.PoolMode - Enum Class in ai.beestream.jitnn
F.Reduction - Enum Class in ai.beestream.jitnn
How a loss reduces its per-element values.
f16(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
The device array of FP16 tensor t.
F16 - Enum constant in enum class ai.beestream.jitnn.DType
IEEE 16-bit float.
F16 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
IEEE 16-bit float.
f32(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
The device array of FP32 tensor t, to pass to a kernel.
F32 - Enum constant in enum class ai.beestream.jitnn.DType
IEEE 32-bit float, the default.
F32 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
32-bit float.
F64 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
64-bit float; raw bytes only.
factor() - Method in record class ai.beestream.jitnn.RopeScaling.Linear
Returns the value of the factor record component.
factor() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Returns the value of the factor record component.
fakeQuantize(Tensor, float, float, int) - Static method in class ai.beestream.jitnn.F
Quantises x to bits (8 or 16) bits and back: clamp(round_half_even(x / scale), lowest, lowest + 2^bits - 1) * scale, lowest being the lowest code minus the zero point (ONNX QuantizeLinear followed by DequantizeLinear).
fattr - Variable in class ai.beestream.jitnn.engine.Node
The float attributes of this application (scales, constants); never null.
featureLogProb() - Method in class ai.beestream.jitnn.ml.NaiveBayes
Multinomial: the per-class log probabilities of the features [classes, features].
fft(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.fft of complex x along the last axis, full length, BACKWARD scaling.
fft(Tensor, int, int, Fft.Norm) - Static method in class ai.beestream.jitnn.Fft
torch.fft.fft of complex x [..., 2]; n > 0 crops or zero-pads axis dim to n.
Fft - Class in ai.beestream.jitnn
Fourier transforms after torch.fft, and the short-time transforms torch.stft and torch.istft.
Fft.Norm - Enum Class in ai.beestream.jitnn
Scaling, torch's norm: BACKWARD scales the inverse by 1/n, ORTHO both by 1/sqrt(n), FORWARD the forward by 1/n.
fft2(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.fft2: the transform over the last two complex axes.
field() - Method in class ai.beestream.jitnn.onnx.Proto
The field number of the current field.
finishAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int) - Method in interface ai.beestream.jitnn.dist.Collective
finishAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int) - Method in class ai.beestream.jitnn.dist.NcclCollective
Waits for the all-reduce started with tag and unpacks it into new tensors.
firstRank() - Method in interface ai.beestream.jitnn.dist.Collective
The global rank of this process's first rank (0 in a single process).
firstRank() - Method in class ai.beestream.jitnn.dist.NcclCollective
 
fit(Tensor) - Method in class ai.beestream.jitnn.ml.KMeans
Fits from k-means++ seeds.
fit(Tensor) - Method in class ai.beestream.jitnn.ml.Pca
Fits samples x [n, features].
fit(Tensor, int[]) - Method in class ai.beestream.jitnn.ml.GradientBoosting
Fits a classifier to samples x [n, features] and labels y.
fit(Tensor, int[]) - Method in class ai.beestream.jitnn.ml.KNeighbors
Classification: integer labels.
fit(Tensor, int[]) - Method in class ai.beestream.jitnn.ml.LinearSvc
Fits samples x [n, features] and labels y.
fit(Tensor, int[]) - Method in class ai.beestream.jitnn.ml.LogisticRegression
Fits samples x [n, features] and labels y.
fit(Tensor, int[]) - Method in class ai.beestream.jitnn.ml.NaiveBayes
Fits samples x [n, features] (counts for the multinomial kind) and labels y.
fit(Tensor, int[]) - Method in class ai.beestream.jitnn.ml.RandomForest
Fits a classifier to samples x [n, features] and labels y.
fit(Tensor, Tensor) - Method in class ai.beestream.jitnn.ml.GradientBoosting
Regression targets [n].
fit(Tensor, Tensor) - Method in class ai.beestream.jitnn.ml.KMeans
Fits from the given initial centres [k, d].
fit(Tensor, Tensor) - Method in class ai.beestream.jitnn.ml.KNeighbors
Regression: targets [n] or [n, t].
fit(Tensor, Tensor) - Method in class ai.beestream.jitnn.ml.LinearRegression
y [n] or [n, t].
fit(Tensor, Tensor) - Method in class ai.beestream.jitnn.ml.RandomForest
Regression targets [n].
fitTransform(Tensor) - Method in class ai.beestream.jitnn.ml.Tsne
The embedding [n, components] of samples x [n, features], starting from their PCA projection.
fixed32() - Method in class ai.beestream.jitnn.onnx.Proto
Reads a little-endian 32-bit value.
fixed64() - Method in class ai.beestream.jitnn.onnx.Proto
Reads a little-endian 64-bit value.
FLASH - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.flash.
FLASH_MMA - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.flash.mma.
flatten() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
flatten(int, int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
Flatten - Class in ai.beestream.jitnn.nn
[N, d1, d2, ...] to [N, d1 * d2 * ...], a view.
Flatten() - Constructor for class ai.beestream.jitnn.nn.Flatten
 
flip(int...) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
flip(Tensor, int...) - Static method in class ai.beestream.jitnn.F
x reversed along each of axes (torch.flip).
FLOAT - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 32-bit float.
FLOAT16 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: IEEE 16-bit float.
float32() - Method in class ai.beestream.jitnn.onnx.Proto
Reads a 32-bit float.
float64() - Method in class ai.beestream.jitnn.onnx.Proto
Reads a 64-bit float.
floats() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the floats record component.
floats() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
The values as floats (FLOAT, FLOAT16, BFLOAT16, DOUBLE and the integer types).
floats(Proto.FloatSink) - Method in class ai.beestream.jitnn.onnx.Proto
Repeated float values: one fixed32, or a packed run.
floats(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
An array of float32 values; throws when the key is absent or holds another type.
floats(String, int...) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Weight name as FP32 values, checking its shape.
floatValue(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
A float32 or float64 value (an integer value is accepted too).
floatValue(String) - Method in class ai.beestream.jitnn.io.StateDict
The value name as a float; throws when it is missing.
floatValue(String, float) - Method in class ai.beestream.jitnn.gguf.GgufFile
GgufFile.floatValue(String), or fallback when the key is absent.
floor() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
floor() - Method in class ai.beestream.jitnn.Tensor
Element-wise floor.
fmax(Tensor) - Method in class ai.beestream.jitnn.Tensor
Element-wise maximum ignoring a NaN operand (torch.fmax).
fmin(Tensor) - Method in class ai.beestream.jitnn.Tensor
Element-wise minimum ignoring a NaN operand (torch.fmin).
fmod(Tensor) - Method in class ai.beestream.jitnn.Tensor
this - trunc(this / o) o, with broadcasting: the remainder with the sign of the dividend (torch.fmod).
fold(Tensor, int[], int[], int[], int[], int[]) - Static method in class ai.beestream.jitnn.F
Folds columns back into an image, summing overlaps (torch fold, ONNX Col2Im); see Spatial.fold.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.AdaptiveAvgPool
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.AdaptiveMaxPool
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Add
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.AvgPool1d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.AvgPool2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.AvgPool3d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.BatchNorm2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.CELU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Conv1d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Conv2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Conv3d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.ConvTranspose2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Cube
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Dropout
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.ELU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Embedding
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.EmbeddingBag
Every index its own bag.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Flatten
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.GlobalAvgPool2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.GraphConv
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.GroupNorm
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.GRU
Output sequence of the last layer, from zero initial states.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.HardSigmoid
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.HardSwish
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.HardTanh
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.InstanceNorm
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.LayerNorm
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.LeakyReLU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Linear
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.LocalResponseNorm
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.LSTM
Output sequence of the last layer, from zero initial states.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.MaxPool1d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.MaxPool2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.MaxPool3d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Mish
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Bert
Hidden states [B, T, dim] for INT32 tokens [B, T], without a padding mask.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.CharRnn
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Dcgan.Discriminator
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Dcgan.Generator
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Doc2Vec
The document vectors of ids x.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.GloVe
The word vectors W + W~ of ids x.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.GPT2
Logits [B, T, vocab] for INT32 tokens [B, T].
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Llama
Logits [B, T, vocab] for INT32 tokens [B, T].
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.UNet
Logits [B, 1, H, W] for x [B, in, H, W] (H and W divisible by 4).
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Vae
Reconstruction with the mean latent (no sampling).
forward(Tensor) - Method in class ai.beestream.jitnn.nn.models.Word2Vec
The word vectors of ids x.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Module
Applies the module to x.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.MultiheadAttention
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Pad
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.PReLU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.ReLU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.ReLU6
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.RMSNorm
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.RReLU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.SELU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Sequential
Applies the layers, fusing a ReLU into the layer before it: Linear + ReLU and Conv2d + ReLU run as one GEMM with the activation in the epilogue, BatchNorm2d + ReLU as one kernel.
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Sigmoid
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Softplus
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Softsign
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.SyncBatchNorm2d
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Tanh
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.ThresholdedReLU
 
forward(Tensor) - Method in class ai.beestream.jitnn.nn.Upsample
 
forward(Tensor, int[]) - Method in class ai.beestream.jitnn.nn.EmbeddingBag
One row per bag: bag b covers indices offsets[b] .. offsets[b + 1] - 1.
forward(Tensor, SparseMatrix.Valued) - Method in class ai.beestream.jitnn.nn.GraphConv
Applies the layer to node features x [nodes, in] over the graph's normalised adjacency (see SparseMatrix.gcn(int, int[], int[])).
forward(Tensor, Tensor, boolean) - Method in class ai.beestream.jitnn.nn.MultiheadAttention
Attention over x [B, T, D] with an optional additive mask (null for none) and optional causal masking.
forward(Tensor, Tensor, int[]) - Method in class ai.beestream.jitnn.nn.GRU
From initial states [layers directions, B, H] (null: zeros), for sequences of lengths (null: all T).
forward(Tensor, Tensor, Tensor, int[]) - Method in class ai.beestream.jitnn.nn.LSTM
From initial states [layers directions, B, H] (null: zeros), for sequences of lengths (null: all T).
FORWARD - Enum constant in enum class ai.beestream.jitnn.Fft.Norm
The forward transform scaled by 1/n, the inverse unscaled.
forwardInternal(ParameterStore, NDList, boolean, PairList) - Method in class ai.beestream.jitnn.djl.JitnnSymbolBlock
 
forwardRelu(Tensor) - Method in class ai.beestream.jitnn.nn.BatchNorm2d
relu(forward(x)), fused into the normalisation kernel.
forwardRelu(Tensor) - Method in class ai.beestream.jitnn.nn.Conv2d
relu(forward(x)), with the bias and ReLU fused into the convolution's output.
forwardRelu(Tensor) - Method in class ai.beestream.jitnn.nn.Linear
relu(forward(x)) as one fused GEMM; Sequential uses it for a Linear followed by ReLU.
fp8(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
FP8 tensors: one byte per value.
FP8_E4M3 - Enum constant in enum class ai.beestream.jitnn.DType
OCP 8-bit float with 4 exponent and 3 mantissa bits, stored without its scale.
FP8_E5M2 - Enum constant in enum class ai.beestream.jitnn.DType
OCP 8-bit float with 5 exponent and 2 mantissa bits, stored without its scale.
from(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
fromCoo(int, int, int[], int[]) - Static method in class ai.beestream.jitnn.SparseMatrix
A rows x cols structure with entries at (rowIndex[e], colIndex[e]); duplicates add up.
fromHuggingFace(Path) - Static method in class ai.beestream.jitnn.nn.models.Bert
A BERT-family encoder from a Hugging Face model directory: config.json and safetensors weights, by model_type: bert (bert-base-uncased and other BERT checkpoints, as BertModel, BertForMaskedLM or BertForPreTraining); roberta (roberta-base and the like): positions counted from the padding index + 1, skipping padding, as RoBERTa does; LayerNorm epsilon 1e-5; distilbert: no token types, no pooler. The MLM head is loaded when the checkpoint has it; otherwise it keeps its initial values.
fromHuggingFace(Path) - Static method in class ai.beestream.jitnn.nn.models.GPT2
GPT-2 from a Hugging Face model directory (gpt2, gpt2-medium, ... or a fine-tuned copy): config.json and safetensors weights, as GPT2Model or GPT2LMHeadModel save them.
fromHuggingFace(Path) - Static method in class ai.beestream.jitnn.nn.models.Llama
A Llama from a Hugging Face checkpoint directory (config.json and safetensors, as HfCheckpoint): model_type llama (or mistral without a sliding window, qwen2 with its q, k, v biases), grouped-query attention, tied or separate output weights, and linear or Llama 3.1 RoPE scaling.
fromId(int) - Static method in enum class ai.beestream.jitnn.gguf.GgmlType
The type with ggml id id, or null for an id this reader does not know.
fromProperties(Properties) - Static method in class ai.beestream.jitnn.JitnnConfig
The options set in properties (jitnn.* keys); other keys are ignored.
fromSystemProperties() - Static method in class ai.beestream.jitnn.JitnnConfig
The options set as system properties now.
full(float, int...) - Static method in class ai.beestream.jitnn.Tensor
An FP32 tensor of the given shape with every element value.
full(Shape, float, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
FUSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.fuse.
FUSE_ELEMENTWISE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.fuse.elementwise.
FUSE_PROGRAM - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.fuse.program.
FUSED_ATTENTION - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.fusedAttention.

G

g() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the g record component.
gather(int, Tensor) - Method in class ai.beestream.jitnn.Tensor
torch.gather along axis with an INT32 index of this shape except along the axis.
gather(NDArray, int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
gatherNd(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
gatherND(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Slices of data picked by INT32 index rows indices [..., k] (ONNX GatherND, batch_dims 0); differentiable in data.
GAUSSIAN - Enum constant in enum class ai.beestream.jitnn.ml.NaiveBayes.Kind
A normal distribution per class and feature (GaussianNB).
gaussianKl(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
KL divergence of N(mu, exp(logvar)) from N(0, 1), summed over every element: -0.5 * sum(1 + logvar - mu^2 - exp(logvar)), the VAE regulariser.
gcn(int, int[], int[]) - Static method in class ai.beestream.jitnn.SparseMatrix
The normalised adjacency of a graph convolutional network (Kipf and Welling, 2017), D^-1/2 (A + I) D^-1/2 for the directed edges src[e] -> dst[e] of an n-node graph (messages flow from source to destination, as in PyTorch Geometric's GCNConv).
ge(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where this >= o, else 0, with broadcasting.
gelu() - Method in class ai.beestream.jitnn.Tensor
GELU, exact form x * Phi(x) (PyTorch's default, approximate='none').
geluTanh() - Method in class ai.beestream.jitnn.Tensor
GELU with the tanh approximation (GPT-2's form, PyTorch's approximate='tanh').
GEMM - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.gemm.
Generator(int, int, long) - Constructor for class ai.beestream.jitnn.nn.models.Dcgan.Generator
A generator from latent vectors of latent values, width channels in its last hidden layer, initialised from seed.
get(JitnnConfig.Option) - Method in class ai.beestream.jitnn.JitnnConfig
The value of option: the one set, else its default (null for a machine-dependent default).
get(NDIndex) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
get(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
The value of key, or null when absent.
getAlternativeEngine() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
getDirectParameters() - Method in class ai.beestream.jitnn.djl.JitnnSymbolBlock
 
getEngine() - Method in class ai.beestream.jitnn.djl.JitnnEngineProvider
 
getEngine() - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
getEngineName() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
getEngineName() - Method in class ai.beestream.jitnn.djl.JitnnEngineProvider
 
getEngineRank() - Method in class ai.beestream.jitnn.djl.JitnnEngineProvider
 
getGpuCount() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
getGradient() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
getInstance() - Static method in class ai.beestream.jitnn.djl.JitnnEngine
The engine, created on first use.
getNDArrayInternal() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
getRank() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
getResourceNDArrays() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
getVersion() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
GgmlType - Enum Class in ai.beestream.jitnn.gguf
The tensor element types a GGUF file can hold (ggml's ggml_type), with their ids, block sizes and the jitNN DType each one loads as.
GGUF_TILED_ROWS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.gguf.tiledRows.
GgufFile - Class in ai.beestream.jitnn.gguf
A GGUF file (llama.cpp's model format, versions 2 and 3): its metadata and its tensors, which load as persistent jitNN tensors.
GgufTensorInfo - Record Class in ai.beestream.jitnn.gguf
One tensor listed in a GGUF file.
GgufTensorInfo(String, long[], int, GgmlType, long, long) - Constructor for record class ai.beestream.jitnn.gguf.GgufTensorInfo
Creates an instance of a GgufTensorInfo record class.
globalAvgPool2d(Tensor) - Static method in class ai.beestream.jitnn.F
Mean over the spatial dimensions: [N, C, H, W] to [N, C].
GlobalAvgPool2d - Class in ai.beestream.jitnn.nn
[N, C, H, W] to [N, C], the mean over each feature map (adaptive average pooling to 1x1, flattened).
GlobalAvgPool2d() - Constructor for class ai.beestream.jitnn.nn.GlobalAvgPool2d
 
GloVe - Class in ai.beestream.jitnn.nn.models
GloVe (Pennington et al., 2014): word and context vectors with biases fitted to the logarithm of the co-occurrence counts, weighted by f(x) = min(1, (x / xMax)^alpha).
GloVe(int, int, float, float, long) - Constructor for class ai.beestream.jitnn.nn.models.GloVe
GloVe over vocab words with vectors of dim and weighting f(x) = min(1, (x / xMax)^alpha), initialised from seed.
GloVe(int, int, long) - Constructor for class ai.beestream.jitnn.nn.models.GloVe
GloVe with the paper's weighting (xMax 100, alpha 3/4).
gloveLoss(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
GloVe's weighted least squares (Pennington et al., 2014) over a batch of co-occurrences: mean of weight (wi . wj + bi + bj - logX)^2, with wi, wj [B, D] and the rest [B].
GPT2 - Class in ai.beestream.jitnn.nn.models
A GPT-2 style decoder: token and learned position embeddings, pre-LayerNorm blocks of causal multi-head attention and a GELU (tanh form) MLP of 4 x dim, a final LayerNorm, and logits from the token embedding (tied weights).
GPT2(int, int, int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.models.GPT2
A decoder over vocab tokens and up to context positions, with layers blocks of dim features in heads heads, initialised from seed.
grad() - Method in class ai.beestream.jitnn.Tensor
The accumulated gradient of a leaf tensor after Tensor.backward(), or null.
Grad - Class in ai.beestream.jitnn
Whether new operators record what autograd needs.
Grad.Scope - Interface in ai.beestream.jitnn
A try-with-resources scope that restores the previous gradient mode.
GradientBoosting - Class in ai.beestream.jitnn.ml
Gradient-boosted trees with second-order (Newton) leaves on binned features, after scikit-learn's HistGradientBoosting and XGBoost: squared error for regression, the logistic loss for two classes and the softmax loss (one tree per class and round) for more.
GradientBoosting(boolean, int, float, int) - Constructor for class ai.beestream.jitnn.ml.GradientBoosting
Boosting with scikit-learn's defaults for the rest: 20 samples per leaf, no L2 penalty, 255 bins.
GradientBoosting(boolean, int, float, int, int, float, int) - Constructor for class ai.beestream.jitnn.ml.GradientBoosting
Boosting for classification or regression: rounds trees of depth at most maxDepth, shrunk by learningRate, with at least minSamplesLeaf samples per leaf, L2 penalty lambda on the leaf values, and features in at most maxBins bins.
GradScaler - Class in ai.beestream.jitnn.nn.optim
Dynamic loss scaling for FP16 training, as PyTorch's torch.amp.GradScaler: the loss is multiplied by a scale before backward; GradScaler.unscale(Optimizer) makes the optimiser divide the gradients by it (inside the update, together with any clipping) and look for overflows on the device; GradScaler.step(Optimizer) runs the optimiser, which skips the update on the device when an overflow was found; GradScaler.update() (after the step has run, e.g. after loss.item()) halves the scale on an overflow, or doubles it after growthInterval clean steps.
GradScaler() - Constructor for class ai.beestream.jitnn.nn.optim.GradScaler
A scaler with PyTorch's defaults: initial scale 65536, growth 2 after 2000 clean steps, backoff 0.5.
GradScaler(float, float, float, int) - Constructor for class ai.beestream.jitnn.nn.optim.GradScaler
A scaler starting at initScale that multiplies the scale by growthFactor after growthInterval steps without overflow and by backoffFactor after an overflow.
graph() - Method in class ai.beestream.jitnn.onnx.OnnxModel
The model's main graph.
graph(SparseMatrix.Valued) - Method in class ai.beestream.jitnn.nn.GraphConv
Sets the graph the next forward calls use.
Graph(String, List, Map, List, List) - Constructor for record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Creates an instance of a Graph record class.
GRAPH_MAX_TASKS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.graph.maxTasks.
GraphConv - Class in ai.beestream.jitnn.nn
Graph convolution (Kipf and Welling, 2017), PyTorch Geometric's GCNConv:  (x W^T) + b, with  the normalised adjacency SparseMatrix.gcn(int, int[], int[]) of the graph, given with each call since one layer serves many graphs. x [nodes, in] gives [nodes, out].
GraphConv(int, int, long) - Constructor for class ai.beestream.jitnn.nn.GraphConv
Graph convolution from in to out features: a linear map without bias initialised from seed, and a bias of zeros.
grid1D(int) - Static method in class ai.beestream.jitnn.engine.Emitter
A 1-D grid of at least n threads, in groups of Emitter.LOCAL.
grid2D(int, int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
A 2-D grid covering x * y threads in tiles of tile * tile.
groupNorm(Tensor, int, Tensor, Tensor, float) - Static method in class ai.beestream.jitnn.F
GroupNorm (Wu and He) of x [N, C, ...] in groups groups of channels: each (sample, group) normalised over its channels and positions, then a per-channel affine (weight and bias may be null, as PyTorch with affine=False).
GroupNorm - Class in ai.beestream.jitnn.nn
GroupNorm over [N, C, ...] with a per-channel affine (weight 1, bias 0 at the start), PyTorch's nn.GroupNorm.
GroupNorm(int, int, float, boolean) - Constructor for class ai.beestream.jitnn.nn.GroupNorm
GroupNorm of channels channels in groups groups, with the given eps, with or without the per-channel affine.
groups(int) - Static method in class ai.beestream.jitnn.engine.Emitter
groups work-groups of Emitter.LOCAL threads each.
gru(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
One GRU layer over x [T, B, I] (time first), PyTorch's equations and gate order (r, z, n): wih [3H, I], whh [3H, H], biases [3H], h0 [B, H].
gru(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor, int[]) - Static method in class ai.beestream.jitnn.F
GRU - Class in ai.beestream.jitnn.nn
A stack of GRU layers over [T, B, I] sequences (time first), optionally bidirectional, PyTorch's nn.GRU equations, gate order, parameter order and initialisation: every weight and bias uniform in [-1/sqrt(H), 1/sqrt(H)).
GRU(int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.GRU
layers GRU layers from input to hidden features, optionally bidirectional, initialised from seed.
GRU(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.GRU
layers unidirectional GRU layers from input to hidden features, initialised from seed.
GRU.Output - Record Class in ai.beestream.jitnn.nn
Output sequence of the last layer [T, B, directions H] and the final hidden states, [layers directions, B, H] each.
gt(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where this > o, else 0, with broadcasting.
gt(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
gt(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
gte(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
gte(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 

H

h() - Method in record class ai.beestream.jitnn.nn.GRU.Output
Returns the value of the h record component.
h() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
Returns the value of the h record component.
half() - Method in class ai.beestream.jitnn.Tensor
This tensor converted to FP16 (to(DType.F16)).
HALF_ATTENTION - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.halfAttention.
halfActivations() - Static method in class ai.beestream.jitnn.Autocast
True inside Autocast.enableHalf(): FP16-capable operators return FP16.
hardSigmoid() - Method in class ai.beestream.jitnn.Tensor
HardSigmoid, PyTorch's form: clamp(x / 6 + 1/2, 0, 1).
HardSigmoid - Class in ai.beestream.jitnn.nn
HardSigmoid: clamp(x / 6 + 1/2, 0, 1).
HardSigmoid() - Constructor for class ai.beestream.jitnn.nn.HardSigmoid
 
hardSwish() - Method in class ai.beestream.jitnn.Tensor
HardSwish: x hardSigmoid(x).
HardSwish - Class in ai.beestream.jitnn.nn
HardSwish: x hardSigmoid(x).
HardSwish() - Constructor for class ai.beestream.jitnn.nn.HardSwish
 
hardTanh(float, float) - Method in class ai.beestream.jitnn.Tensor
HardTanh: x clamped to [lo, hi] (PyTorch's default -1, 1).
HardTanh - Class in ai.beestream.jitnn.nn
HardTanh: x clamped to [min, max].
HardTanh(float, float) - Constructor for class ai.beestream.jitnn.nn.HardTanh
Clamps to [min, max] (PyTorch's default -1, 1).
has(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
Whether the metadata has key.
has(String) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
True when the checkpoint holds weight name (under any accepted spelling).
hasCapability(String) - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
hasGradient() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
hashCode() - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.io.SafeTensors.Header
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
Returns a hash code value for this object.
hashCode() - Method in class ai.beestream.jitnn.JitnnConfig
 
hashCode() - Method in record class ai.beestream.jitnn.nn.GRU.Output
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.nn.models.Vae.Output
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.RopeScaling.Linear
Returns a hash code value for this object.
hashCode() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Returns a hash code value for this object.
hashCode() - Method in class ai.beestream.jitnn.Shape
 
hashCode() - Method in record class ai.beestream.jitnn.SparseMatrix.Valued
Returns a hash code value for this object.
hasPooler() - Method in class ai.beestream.jitnn.nn.models.Bert
True when the model has the pooler (a Hugging Face checkpoint without one, such as roberta-base, gives none).
hasTensor(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
Whether the file lists a tensor named name.
hasTensor(String) - Method in class ai.beestream.jitnn.io.StateDict
Whether the dictionary holds a tensor named name.
Header(Map, Map) - Constructor for record class ai.beestream.jitnn.io.SafeTensors.Header
Creates an instance of a Header record class.
HfCheckpoint - Class in ai.beestream.jitnn.nn.models
A Hugging Face model directory: config.json and the weights in model.safetensors or in shards listed by model.safetensors.index.json.
hidden() - Method in class ai.beestream.jitnn.nn.GRU
The hidden size per direction.
hidden() - Method in class ai.beestream.jitnn.nn.LSTM
The hidden size per direction.
highFreqFactor() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Returns the value of the highFreqFactor record component.
hingeLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Hinge loss with targets in {-1, 1}: max(0, 1 - target * input), Keras's hinge (mean).
HOST_POOL_BYTES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.hostPool.bytes.
HostCollective - Class in ai.beestream.jitnn.dist
Collectives staged through host memory: each rank copies its tensor to the host, the reduction runs on the host in rank order (so every rank gets bit-identical results, run after run), and each rank copies the result back.
HostCollective(int) - Constructor for class ai.beestream.jitnn.dist.HostCollective
Collectives among worldSize ranks of this process.
huberLoss(Tensor, Tensor, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Huber loss with threshold delta: 0.5 d^2 where |d| < delta, else delta (|d| - delta / 2), d = input - target (PyTorch's huber_loss).

I

i() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the i record component.
I16 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
16-bit signed integer; raw bytes only.
i32(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
The device array of INT32 tensor t.
I32 - Enum constant in enum class ai.beestream.jitnn.DType
32-bit signed integer, for labels and indices.
I32 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
32-bit signed integer.
I64 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
64-bit signed integer; raw bytes only.
I8 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
8-bit signed integer; raw bytes only.
iattr - Variable in class ai.beestream.jitnn.engine.Node
The integer attributes of this application (shapes, axes, flags); never null.
id() - Method in enum class ai.beestream.jitnn.gguf.GgmlType
ggml's numeric id for this type, as stored in the file.
id(String) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
The id of token, or -1.
id(String) - Method in class ai.beestream.jitnn.nn.text.Vocabulary
The id of word, or -1 when it is not in the vocabulary.
idArray() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns the value of the idArray record component.
ids() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
The token ids as an INT32 tensor [batch, length].
ids() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Returns the value of the ids record component.
ifft(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.ifft of complex x along the last axis, full length, BACKWARD scaling (1/n).
ifft(Tensor, int, int, Fft.Norm) - Static method in class ai.beestream.jitnn.Fft
torch.fft.ifft.
ifft2(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.ifft2.
images - Variable in class ai.beestream.jitnn.data.Cifar10
The images [count, 3, 32, 32], row-major.
images - Variable in class ai.beestream.jitnn.data.Mnist
The images [count, 28 * 28], row-major.
in(int) - Method in class ai.beestream.jitnn.engine.Node
Input i.
index() - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Returns the value of the index record component.
indexSelect(int, Tensor) - Method in class ai.beestream.jitnn.Tensor
Elements idx[0 .. m-1] (INT32 [m]) of axis, torch.index_select.
inertia() - Method in class ai.beestream.jitnn.ml.KMeans
Sum of squared distances of the training samples to their nearest centre.
inference(Supplier) - Static method in class ai.beestream.jitnn.Grad
Runs a forward pass for inference: no gradient recording, and every tensor created inside except the returned one is internal, so its buffer is reused as soon as the computation no longer needs it.
inferenceScope() - Static method in class ai.beestream.jitnn.Grad
The scope form of Grad.inference(Supplier), for code that computes several outputs: no gradients, and intermediates are internal (their buffers are reused and inference fusions apply).
info(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
The listing of tensor name: its type, shape and offset; throws when there is none.
initializers() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns the value of the initializers record component.
input(int) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Input k, or "" when absent (ONNX marks skipped optional inputs with "").
inputNames() - Method in class ai.beestream.jitnn.onnx.OnnxSession
The names of the graph's inputs, in graph order.
inputNames() - Method in class ai.beestream.jitnn.tf.TfModel
The model's input names, as run takes them.
inputs - Variable in class ai.beestream.jitnn.engine.Node
The operator's inputs.
inputs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns the value of the inputs record component.
inputs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns the value of the inputs record component.
inputs(DType...) - Method in class ai.beestream.jitnn.CustomOp.Builder
The inputs' element types, in order (null accepts any type).
install(JitnnConfig) - Static method in class ai.beestream.jitnn.JitnnConfig
Makes config the runtime's configuration.
instanceNorm(Tensor, Tensor, Tensor, float) - Static method in class ai.beestream.jitnn.F
InstanceNorm: GroupNorm with one channel per group (affine when weight and bias are given).
InstanceNorm - Class in ai.beestream.jitnn.nn
InstanceNorm over [N, C, ...]: each sample and channel normalised over its positions, PyTorch's nn.InstanceNorm1d/2d/3d without running statistics; affine optional (PyTorch's default: none).
InstanceNorm(int, float, boolean) - Constructor for class ai.beestream.jitnn.nn.InstanceNorm
InstanceNorm of channels channels with the given eps, with or without a per-channel affine (weight 1, bias 0 at the start).
INT16 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 16-bit signed integer.
INT32 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 32-bit signed integer.
INT64 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 64-bit signed integer.
INT8 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 8-bit signed integer.
integer(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
An integer value of any width that fits in an int.
integer(String, int) - Method in class ai.beestream.jitnn.gguf.GgufFile
GgufFile.integer(String), or fallback when the key is absent.
intercept() - Method in class ai.beestream.jitnn.ml.LinearRegression
The fitted intercept (zero without one).
intercept() - Method in class ai.beestream.jitnn.ml.LinearSvc
The intercepts, one per row of LinearSvc.coef().
intercept() - Method in class ai.beestream.jitnn.ml.LogisticRegression
The intercepts, one per row of the weights.
intern(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
interpolate(Tensor, int, int, boolean, boolean) - Static method in class ai.beestream.jitnn.F
Resizes x [N, C, H, W] to [N, C, height, width], nearest or bilinear (PyTorch's interpolate with size).
ints() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the ints record component.
ints(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
An array of 8-, 16- or 32-bit integers (other than uint32).
intValue(String) - Method in class ai.beestream.jitnn.io.StateDict
The value name as an int; throws when it is missing.
inv(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Inverse of square matrices.
inverseTransform(Tensor) - Method in class ai.beestream.jitnn.ml.Pca
Back from the projection to the feature space.
irfft(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.irfft along the last axis: a real signal of length 2 (m - 1), BACKWARD scaling (1/n).
irfft(Tensor, int, int, Fft.Norm) - Static method in class ai.beestream.jitnn.Fft
torch.fft.irfft: the real signal of length n (default 2 (m - 1)) whose spectrum's non-negative half is x [..., m, 2]; the imaginary parts at frequency 0 and n / 2 are ignored.
isBlockQuantized() - Method in enum class ai.beestream.jitnn.DType
True for GGUF's block-quantised formats.
isClosed() - Method in class ai.beestream.jitnn.nn.Module
Whether Module.close() was called.
isCuda() - Method in class ai.beestream.jitnn.engine.Emitter
True when the plan runs on CUDA.
isEnabled() - Static method in class ai.beestream.jitnn.Grad
Whether operators recorded now on this thread record what autograd needs.
isExecuted() - Method in class ai.beestream.jitnn.engine.Node
Whether the node has run (always true for an untraced node).
isFiniteMask() - Method in class ai.beestream.jitnn.Tensor
1 where finite, else 0.
isFloat() - Method in enum class ai.beestream.jitnn.DType
True for the floating-point types.
isInf() - Method in class ai.beestream.jitnn.Tensor
1 where +inf or -inf, else 0.
isInfinite() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
isLeaf() - Method in class ai.beestream.jitnn.Tensor
True unless the tensor is the output of an operator that tracks gradients (it is a leaf of autograd).
isMetal() - Method in class ai.beestream.jitnn.engine.Emitter
True when the plan runs on Metal (simdgroup matrix kernels are available).
isNaN() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
isNaN() - Method in class ai.beestream.jitnn.Tensor
1 where NaN, else 0 (tested by bit pattern).
isPersistent() - Method in class ai.beestream.jitnn.Tensor
True for persistent device storage: weights (Tensor.persistent(Tensor), Tensor.parameter(Tensor)) and their views.
isPresent(Path) - Static method in class ai.beestream.jitnn.data.Cifar10
Whether dir, or its cifar-10-batches-bin subdirectory, holds the CIFAR-10 binary files.
isPresent(Path) - Static method in class ai.beestream.jitnn.data.Mnist
Whether dir holds the MNIST training images (plain or gzipped).
isSet(JitnnConfig.Option) - Method in class ai.beestream.jitnn.JitnnConfig
Whether option was set (as a system property or with the builder) rather than left at its default.
istft(Tensor, int, int, Tensor, boolean, boolean, int) - Static method in class ai.beestream.jitnn.Fft
torch.istft(x, nFft, hop, window=window, center, normalized, onesided=True, length): x [(B,) nFft / 2 + 1, frames, 2] back to [(B,) length] (length <= 0: as long as the frames reach, less the centre padding), by overlap-add divided by the summed squared window.
isTraining() - Method in class ai.beestream.jitnn.nn.Module
Whether the module is in training mode (the default).
item() - Method in class ai.beestream.jitnn.Tensor
The value of a single-element tensor, as a float.
iterations() - Method in class ai.beestream.jitnn.ml.KMeans
The iterations the kept run took.

J

JitnnConfig - Class in ai.beestream.jitnn
jitNN's options, in one immutable object.
JitnnConfig.Builder - Class in ai.beestream.jitnn
Builds a configuration option by option.
JitnnConfig.Option<T> - Class in ai.beestream.jitnn
One option: its key, type, default and description.
JitnnConfig.Section - Enum Class in ai.beestream.jitnn
The groups options are listed in.
JitnnEngine - Class in ai.beestream.jitnn.djl
A DJL engine that runs DJL code on jitNN: NDArray operators become lazy jitNN operators, compiled into one TornadoVM plan per step and replayed; DJL's blocks, losses, optimisers and trainers work on top, and Model.load reads ONNX files through jitnn-onnx.
JitnnEngineProvider - Class in ai.beestream.jitnn.djl
Registers JitnnEngine with DJL (through META-INF/services).
JitnnEngineProvider() - Constructor for class ai.beestream.jitnn.djl.JitnnEngineProvider
 
JitnnException - Exception Class in ai.beestream.jitnn
A failure of the device or the runtime: a plan that TornadoVM could not compile or run, a vendor library call that failed, a native library of the CPU backend that could not be called.
JitnnException(String) - Constructor for exception class ai.beestream.jitnn.JitnnException
A failure described by message.
JitnnException(String, Throwable) - Constructor for exception class ai.beestream.jitnn.JitnnException
A failure described by message, caused by cause.
JitnnModel - Class in ai.beestream.jitnn.djl
A DJL model on jitNN.
JitnnNDArray - Class in ai.beestream.jitnn.djl
A DJL NDArray backed by a lazy jitNN tensor.
JitnnNDManager - Class in ai.beestream.jitnn.djl
Creates JitnnNDArrays.
JitnnRuntime - Class in ai.beestream.jitnn
jitNN's runtime: where operators run, sync points, determinism, profiling and statistics.
JitnnSymbolBlock - Class in ai.beestream.jitnn.djl
An ONNX model loaded by jitnn-onnx, as a DJL block.
jointLogLikelihood(Tensor) - Method in class ai.beestream.jitnn.ml.NaiveBayes
Joint log-likelihood [m, c].
Json - Class in ai.beestream.jitnn.io
A small JSON reader for the files jitNN loads (safetensors headers, Hugging Face configs and tokenizers): objects become LinkedHashMaps in file order, arrays Lists, integers Long, other numbers Double.

K

keep(Tensor) - Static method in class ai.beestream.jitnn.Grad
Marks t as read by the caller: its buffer is never reused.
keepHalf() - Static method in class ai.beestream.jitnn.Autocast
True when FP16 activations should stay FP16 through operators that compute in FP32 (inside Autocast.enableHalf(), outside the backward pass): their results are cast back to FP16, so an FP32 fallback does not turn the rest of the forward pass FP32.
kernelBreakdown() - Static method in class ai.beestream.jitnn.JitnnRuntime
With jitnn.profile=true: device time per operator task ("op#k", the k-th task of the operator) as [launches, nanoseconds], accumulated since the last call, which resets it.
KERNELS - Enum constant in enum class ai.beestream.jitnn.JitnnConfig.Section
jitNN's kernel choices and graph rewrites.
key() - Method in class ai.beestream.jitnn.JitnnConfig.Option
The system property, e.g.
keys() - Method in class ai.beestream.jitnn.gguf.GgufFile
The metadata keys, in file order.
KMeans - Class in ai.beestream.jitnn.ml
k-means clustering with Lloyd's algorithm, after scikit-learn's KMeans: k-means++ seeding (or given initial centres), nInit restarts keeping the lowest inertia, and stopping when the assignments no longer change or the centres move less than tol times the mean feature variance.
KMeans(int, int, float, int, long) - Constructor for class ai.beestream.jitnn.ml.KMeans
k-means with k clusters, at most maxIter iterations per run, tolerance tol, nInit runs, seeding from seed.
KMeans(int, long) - Constructor for class ai.beestream.jitnn.ml.KMeans
k-means with scikit-learn's defaults: 300 iterations, tolerance 1e-4, one run.
kneighbors(Tensor) - Method in class ai.beestream.jitnn.ml.KNeighbors
INT32 indices [m, k] of the k nearest training rows of each query row.
KNeighbors - Class in ai.beestream.jitnn.ml
k-nearest neighbours by brute force on the device (Euclidean distance, uniform weights), after scikit-learn's KNeighborsClassifier and KNeighborsRegressor.
KNeighbors(int) - Constructor for class ai.beestream.jitnn.ml.KNeighbors
Neighbours by the k nearest training samples.

L

l1Loss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Mean absolute error (L1), PyTorch's l1_loss.
labels - Variable in class ai.beestream.jitnn.data.Cifar10
The class of each image.
labels - Variable in class ai.beestream.jitnn.data.Mnist
The digit of each image.
Lamb - Class in ai.beestream.jitnn.nn.optim
LAMB (You et al., 2020): Adam's bias-corrected update plus decoupled weight decay, scaled per parameter by the trust ratio |p| / |update|.
Lamb(List, float, float, float, float, float) - Constructor for class ai.beestream.jitnn.nn.optim.Lamb
LAMB with the given betas, eps and decoupled weight decay.
Lars - Class in ai.beestream.jitnn.nn.optim
LARS (You et al., 2017) with momentum: a per-parameter learning rate eta |p| / (|g| + weightDecay |p|), applied to the decayed gradient and accumulated in a momentum buffer.
Lars(List, float, float, float, float, float) - Constructor for class ai.beestream.jitnn.nn.optim.Lars
LARS with the given momentum, weight decay, trust coefficient eta and eps.
lastIntermediateBytes() - Static method in class ai.beestream.jitnn.JitnnRuntime
Device memory for intermediates (activations, gradients and other temporaries, after buffer reuse) of the last plan this thread ran, in bytes; 0 on the reference backend.
lastProfile() - Static method in class ai.beestream.jitnn.JitnnRuntime
With jitnn.profile=true: [device kernel ns, TornadoVM total ns] of this thread's last plan execution, from TornadoVM's profiler.
latent() - Method in class ai.beestream.jitnn.nn.models.Vae
The size of the latent vector.
launch(CustomOp.Launcher) - Method in class ai.beestream.jitnn.CustomOp.Builder
The device tasks: called once per plan, with the arrays bound to the node's tensors.
layerNorm(Tensor, Tensor, Tensor, float) - Static method in class ai.beestream.jitnn.F
LayerNorm over the last axis: (x - mean) / sqrt(var + eps) * weight + bias (bias may be null).
LayerNorm - Class in ai.beestream.jitnn.nn
LayerNorm over the last axis, with weight 1 and bias 0 initially, like PyTorch's nn.LayerNorm.
LayerNorm(int) - Constructor for class ai.beestream.jitnn.nn.LayerNorm
LayerNorm over features values, with eps 1e-5 and a bias.
LayerNorm(int, float, boolean) - Constructor for class ai.beestream.jitnn.nn.LayerNorm
LayerNorm over features values with the given eps, with or without a bias.
le(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where this <= o, else 0, with broadcasting.
leakyRelu(float) - Method in class ai.beestream.jitnn.Tensor
LeakyReLU: x for x > 0, else slope * x (PyTorch's default slope is 0.01).
LeakyReLU - Class in ai.beestream.jitnn.nn
LeakyReLU with a fixed negative slope (PyTorch's default 0.01).
LeakyReLU() - Constructor for class ai.beestream.jitnn.nn.LeakyReLU
LeakyReLU with slope 0.01.
LeakyReLU(float) - Constructor for class ai.beestream.jitnn.nn.LeakyReLU
LeakyReLU with the given negative slope.
learningRate() - Method in class ai.beestream.jitnn.nn.optim.Optimizer
The current learning rate.
length() - Method in class ai.beestream.jitnn.data.TextCorpus
The number of tokens.
length() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns the value of the length record component.
library(Native.Lib, String, Emitter.TaskAdder) - Method in class ai.beestream.jitnn.engine.Emitter
Adds a TornadoVM library task (a cuBLAS, cuDNN, ... or MLX call, through TaskGraph.libraryTask): it has no launch grid, and runs on the same device and stream as the kernels around it.
libraryForked(Native.Lib, String, Emitter.TaskAdder, String, Emitter.TaskAdder) - Method in class ai.beestream.jitnn.engine.Emitter
A library task that starts work on a stream of its own, joined back by the task Emitter.libraryJoin(Native.Lib, String, Emitter.TaskAdder, String) adds with the same key.
libraryJoin(Native.Lib, String, Emitter.TaskAdder, String) - Method in class ai.beestream.jitnn.engine.Emitter
Linalg - Class in ai.beestream.jitnn
Linear algebra on matrices and batches of matrices [..., rows, cols], after torch.linalg: LU, triangular and general solves, inverse, determinants, Cholesky, QR, symmetric eigendecomposition and SVD.
linear(double) - Static method in interface ai.beestream.jitnn.RopeScaling
Positions divided by factor (rope_type linear).
linear(Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
y = x W^T + b, with W stored [out, in] and b of shape [out] (b may be null); one fused GEMM. x may have any leading dimensions: [..., in] gives [..., out].
Linear - Class in ai.beestream.jitnn.nn
y = x W^T + b with W of shape [out, in].
Linear(double) - Constructor for record class ai.beestream.jitnn.RopeScaling.Linear
Creates an instance of a Linear record class.
Linear(int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Linear
A linear layer from in to out features, with or without a bias, initialised from seed.
Linear(int, int, long) - Constructor for class ai.beestream.jitnn.nn.Linear
A linear layer from in to out features with a bias, initialised from seed.
linearQ8(Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
y = x W^T + b with Q8_0 weights W [out, in]; the gradient reaches x, not W.
LinearRegression - Class in ai.beestream.jitnn.ml
Least squares with an optional L2 penalty alpha (ridge regression), solved in closed form from the normal equations (X^T X + alpha I) w = X^T y on centred data, after scikit-learn's LinearRegression (alpha 0) and Ridge.
LinearRegression() - Constructor for class ai.beestream.jitnn.ml.LinearRegression
Least squares with an intercept.
LinearRegression(float, boolean) - Constructor for class ai.beestream.jitnn.ml.LinearRegression
Ridge regression with penalty alpha (0 for plain least squares), with or without an intercept.
linearRelu(Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
relu(x W^T + b), with the bias and ReLU fused into the GEMM epilogue.
LinearSvc - Class in ai.beestream.jitnn.ml
Linear support vector classification with the squared hinge loss, one-vs-rest, after scikit-learn's LinearSVC (liblinear, intercept_scaling 1): for each class it minimises ||[w, b]||^2 / 2 + C sum_i max(0, 1 - y_i (w . x_i + b))^2, the intercept penalised like a weight, with L-BFGS (the objective is smooth).
LinearSvc() - Constructor for class ai.beestream.jitnn.ml.LinearSvc
A classifier with scikit-learn's defaults: C 1, 1000 iterations.
LinearSvc(float, int) - Constructor for class ai.beestream.jitnn.ml.LinearSvc
A classifier with penalty c and at most maxIter iterations.
linspace(float, float, int, boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
Llama - Class in ai.beestream.jitnn.nn.models
A Llama-style decoder: token embedding, pre-RMSNorm blocks of causal attention with rotary position embedding (rotate-half form) and a SwiGLU feed-forward, a final RMSNorm and an untied output projection; no biases.
Llama(int, int, int, int, int, float, long) - Constructor for class ai.beestream.jitnn.nn.models.Llama
A decoder over vocab tokens with layers blocks of dim features in heads heads, a feed-forward of hidden features and rotary base ropeBase, initialised from seed.
Llama(int, int, int, int, int, int, float, float, boolean, boolean, long) - Constructor for class ai.beestream.jitnn.nn.models.Llama
A Llama with kvHeads key/value heads (grouped-query attention when fewer than heads), RMSNorm epsilon eps, the output projection tied to the token embedding when tied, and biases on the q, k and v projections when qkvBias.
llama3(double, double, double, int) - Static method in interface ai.beestream.jitnn.RopeScaling
Llama 3.1's rescaling (rope_type llama3): frequencies with wavelengths above originalContext / lowFreqFactor are divided by factor, those below originalContext / highFreqFactor are kept, and the band between is interpolated.
Llama3(double, double, double, int) - Constructor for record class ai.beestream.jitnn.RopeScaling.Llama3
Creates an instance of a Llama3 record class.
load(Tensor, String) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Copies weight name into parameter target of the same shape.
load(Path) - Static method in class ai.beestream.jitnn.data.TextCorpus
The bytes of file as a corpus.
load(Path) - Static method in class ai.beestream.jitnn.io.StateDict
Opens a safetensors file; tensors are read when copied in.
load(Path) - Static method in class ai.beestream.jitnn.nn.text.Tokenizer
Loads tokenizer.json, or the one inside directory path.
load(Path) - Static method in class ai.beestream.jitnn.onnx.OnnxModel
Reads an ONNX model file; initializers kept in external data files next to it are memory-mapped.
load(Path) - Static method in class ai.beestream.jitnn.onnx.OnnxSession
Loads an ONNX file and prepares it to run in FP32; external data files next to it are memory-mapped.
load(Path) - Static method in class ai.beestream.jitnn.tf.TfModel
Loads a SavedModel directory's serving_default signature, or a frozen GraphDef file.
load(Path, boolean) - Static method in class ai.beestream.jitnn.data.Cifar10
dir is the directory holding cifar-10-batches-bin (or that directory itself).
load(Path, boolean) - Static method in class ai.beestream.jitnn.data.Mnist
Pixels normalised with MNIST's mean and standard deviation (0.1307, 0.3081).
load(Path, boolean) - Static method in class ai.beestream.jitnn.onnx.OnnxSession
With halfPrecision, activations stay FP16 between operators (products accumulate in FP32) and the linear layers' weights are stored in FP16; outputs are FP32.
load(Path, boolean, boolean) - Static method in class ai.beestream.jitnn.data.Mnist
With unit, pixels are raw / 255 in [0, 1] (targets for generative models); otherwise normalised.
load(Path, String, boolean) - Static method in class ai.beestream.jitnn.tf.TfModel
Loads signature signature of a SavedModel (ignored for a frozen GraphDef, whose inputs are its placeholders and whose outputs are the nodes nothing consumes).
load(Path, String, Map) - Method in class ai.beestream.jitnn.djl.JitnnModel
 
loadConcatenated(Tensor, String...) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Copies the concatenation, along the first axis, of names into target.
loadScalars(StateDict) - Method in class ai.beestream.jitnn.nn.optim.NAdam
 
loadScalars(StateDict) - Method in class ai.beestream.jitnn.nn.optim.Optimizer
loadStateDict(StateDict) - Method in class ai.beestream.jitnn.nn.Module
Overwrites the parameters and buffers with those of d, in place, so compiled plans keep using them.
loadStateDict(StateDict) - Method in class ai.beestream.jitnn.nn.optim.GradScaler
Restores the scale and step count saved by GradScaler.stateDict().
loadStateDict(StateDict) - Method in class ai.beestream.jitnn.nn.optim.LRScheduler
Restores the count saved by LRScheduler.stateDict().
loadStateDict(StateDict) - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Restores Optimizer.stateDict() into this optimiser, in place; type and state shapes must match.
loadTransposed(Tensor, String) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Copies the transpose of 2-D weight name into target (GPT-2's Conv1D layers store [in, out]).
LOCAL - Static variable in class ai.beestream.jitnn.engine.Emitter
Work-group size for 1-D kernels and group reductions.
localRanks() - Method in interface ai.beestream.jitnn.dist.Collective
The number of ranks in this process.
localRanks() - Method in class ai.beestream.jitnn.dist.NcclCollective
 
localResponseNorm(Tensor, int, float, float, float) - Static method in class ai.beestream.jitnn.F
Local response normalisation across channels, PyTorch's local_response_norm (k 1, alpha 1e-4, beta 0.75 by default).
LocalResponseNorm - Class in ai.beestream.jitnn.nn
Local response normalisation across channels, PyTorch's nn.LocalResponseNorm.
LocalResponseNorm(int, float, float, float) - Constructor for class ai.beestream.jitnn.nn.LocalResponseNorm
Normalisation over size neighbouring channels: x / (k + alpha / size * sum x^2)^beta.
log() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
log() - Method in class ai.beestream.jitnn.Tensor
The natural logarithm, element-wise.
log10() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
log1p() - Method in class ai.beestream.jitnn.Tensor
Element-wise log(1 + x), accurate near 0.
log2() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
logicalAnd(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where both are non-zero, with broadcasting.
logicalAnd(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
logicalNot() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
logicalNot() - Method in class ai.beestream.jitnn.Tensor
Element-wise logical not: 1 where x is 0, else 0.
logicalOr(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where either is non-zero, with broadcasting.
logicalOr(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
logicalXor(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where exactly one is non-zero, with broadcasting.
logicalXor(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
LogisticRegression - Class in ai.beestream.jitnn.ml
L2-regularised logistic regression fitted with L-BFGS, after scikit-learn's LogisticRegression (lbfgs): it minimises C sum_i loss_i + ||W||^2 / 2 with the intercept unpenalised; a sigmoid for two classes, the multinomial (softmax) loss for more.
LogisticRegression() - Constructor for class ai.beestream.jitnn.ml.LogisticRegression
A classifier with C 1 and 500 iterations.
LogisticRegression(float, int) - Constructor for class ai.beestream.jitnn.ml.LogisticRegression
A classifier with inverse regularisation strength c and at most maxIter L-BFGS iterations.
logSigmoid(Tensor) - Static method in class ai.beestream.jitnn.F
log sigmoid(x) = -softplus(-x), without overflow.
logSoftmax() - Method in class ai.beestream.jitnn.Tensor
Log-softmax along the last axis.
logSoftmax(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
logvar() - Method in record class ai.beestream.jitnn.nn.models.Vae.Output
Returns the value of the logvar record component.
longs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
The values as longs (INT64, INT32, INT8, UINT8, BOOL, and integral floats).
longs(Proto.LongSink) - Method in class ai.beestream.jitnn.onnx.Proto
Repeated int64 values: one varint, or a packed run.
longs(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
An array of uint32 or 64-bit integers.
longValue(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
An integer value of any width.
longValue(String) - Method in class ai.beestream.jitnn.io.StateDict
The value name as a long; throws when it is missing.
loss(int[], int[], float[]) - Method in class ai.beestream.jitnn.nn.models.GloVe
The loss of a batch of co-occurrences x at (rows, cols), all > 0.
loss(int[], int[], int[], int) - Method in class ai.beestream.jitnn.nn.models.Word2Vec
The loss of a batch: skip-gram takes centres [B] and contexts [B]; CBOW takes the centre words [B] and their contexts [B * window2] (window2 per sample). negatives holds k words per sample.
loss(int[], int[], int[], int[], int) - Method in class ai.beestream.jitnn.nn.models.Doc2Vec
The loss of a batch: document ids [B], the predicted words [B], for PV-DM the context words [B * window2] (null for PV-DBOW), and k negatives per sample.
loss(Vae.Output, Tensor) - Static method in class ai.beestream.jitnn.nn.models.Vae
The negative ELBO per sample: (binary cross-entropy summed over pixels + KL) / B.
lowFreqFactor() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Returns the value of the lowFreqFactor record component.
LP - Enum constant in enum class ai.beestream.jitnn.F.PoolMode
 
lpNormalize(Tensor, int, int, float) - Static method in class ai.beestream.jitnn.F
x / max(||x||_p, eps) along axis, p = 1 or 2 (torch.nn.functional.normalize; ONNX LpNormalization with eps 0).
lr - Variable in class ai.beestream.jitnn.nn.optim.Optimizer
The learning rate.
LRScheduler - Class in ai.beestream.jitnn.nn.optim
Sets an optimiser's learning rate per epoch; call LRScheduler.step() once at the end of each epoch, as with PyTorch's schedulers.
LRScheduler(Optimizer) - Constructor for class ai.beestream.jitnn.nn.optim.LRScheduler
A schedule for optimizer, starting from its current learning rate.
lstm(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
One LSTM layer over x [T, B, I] (time first), PyTorch's equations and gate order (i, f, g, o): wih [4H, I], whh [4H, H], biases [4H], h0 and c0 [B, H].
lstm(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor, Tensor, int[]) - Static method in class ai.beestream.jitnn.F
As F.lstm(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor, Tensor) for padded batches: INT32 lengths [B] (null: all T); outputs past a sequence's length are 0 and h_n, c_n are each sequence's last valid state.
LSTM - Class in ai.beestream.jitnn.nn
A stack of LSTM layers over [T, B, I] sequences (time first), optionally bidirectional, PyTorch's nn.LSTM equations, gate order, parameter order and initialisation: every weight and bias uniform in [-1/sqrt(H), 1/sqrt(H)).
LSTM(int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.LSTM
layers LSTM layers from input to hidden features, optionally bidirectional, initialised from seed.
LSTM(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.LSTM
layers unidirectional LSTM layers from input to hidden features, initialised from seed.
LSTM.Output - Record Class in ai.beestream.jitnn.nn
Output sequence of the last layer [T, B, directions H] and the final hidden states and cell states, [layers directions, B, H] each.
lt(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where this < o, else 0, with broadcasting.
lt(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
lt(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
lte(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
lte(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
lu(Tensor) - Static method in class ai.beestream.jitnn.Linalg
P, L, U with A = P L U, torch.linalg.lu for square matrices.
luFactor(Tensor) - Static method in class ai.beestream.jitnn.Linalg
LU factorisation with partial pivoting, torch.linalg.lu_factor: the packed factors (L unit lower below the diagonal, U on and above) and INT32 row indices perm, with row i of P^T A being row perm[i] of A (0-based, unlike LAPACK's swap list).

M

mapeLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Mean absolute percentage error: 100 |target - input| / max(|target|, eps) (Keras's MAPE, eps 1e-7).
marginRankingLoss(Tensor, Tensor, Tensor, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Margin ranking loss: max(0, -target (x1 - x2) + margin), target 1 or -1.
markdownTable(JitnnConfig.Section) - Static method in class ai.beestream.jitnn.JitnnConfig
The Markdown table of a section's options (property, default, meaning), as docs/CONFIGURATION.md shows it.
masked(Buckets, Buckets, Tensor, Tensor, int, BinaryOperator) - Static method in class ai.beestream.jitnn.nn.Bucketing
An encoder with an additive attention mask (BERT): model takes tokens [B, T] and a mask [B, 1, 1, T] or [B, 1, T, T] (mask null for none).
MASKED - Static variable in class ai.beestream.jitnn.nn.Bucketing
The additive mask value for padded positions.
maskedSelect(Tensor) - Method in class ai.beestream.jitnn.Tensor
The elements where mask (broadcast) is not zero, as a vector (torch.masked_select); reads the count back to the host.
matmul(Tensor) - Method in class ai.beestream.jitnn.Tensor
The matrix product this b: [M, K] x [K, N] to [M, N], or batched over leading axes, which must match unless one side is a plain matrix shared by every batch.
matmul(Tensor, boolean, boolean) - Method in class ai.beestream.jitnn.Tensor
op(this) * op(b), where op transposes when the flag is set, without materialising the transpose.
matMul(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
matrix() - Method in record class ai.beestream.jitnn.SparseMatrix.Valued
Returns the value of the matrix record component.
max() - Method in class ai.beestream.jitnn.Buckets
The largest size.
max() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
max(int[], boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
max(int, boolean) - Method in class ai.beestream.jitnn.Tensor
The maximum along axis, which is kept with size 1 when keepDim is set; Tensor.argmax(int) gives its index.
MAX - Enum constant in enum class ai.beestream.jitnn.dist.Collective.ReduceOp
The maximum over ranks.
MAX - Enum constant in enum class ai.beestream.jitnn.F.BagMode
The element-wise maximum of the rows.
MAX - Enum constant in enum class ai.beestream.jitnn.F.PoolMode
 
maximum(Tensor) - Method in class ai.beestream.jitnn.Tensor
The element-wise maximum of this and o, with broadcasting (Tensor.maximumNaN(Tensor) propagates NaN).
maximum(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
maximum(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
maximumNaN(Tensor) - Method in class ai.beestream.jitnn.Tensor
Element-wise maximum propagating NaN (torch.maximum).
maxPool1d(Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
Max pooling of x [N, C, L] over windows of kernel with the given stride and padding (padding never wins), PyTorch's max_pool1d.
MaxPool1d - Class in ai.beestream.jitnn.nn
Max pooling over 1-D windows, PyTorch's nn.MaxPool1d.
MaxPool1d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.MaxPool1d
Pools windows of kernel with the given stride and padding.
maxPool2d(Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
Max pooling of x [N, C, H, W] over square windows of kernel with the given stride and padding, PyTorch's max_pool2d.
MaxPool2d - Class in ai.beestream.jitnn.nn
Max pooling over square windows of NCHW input, PyTorch's nn.MaxPool2d.
MaxPool2d(int) - Constructor for class ai.beestream.jitnn.nn.MaxPool2d
Pools non-overlapping kernel x kernel windows (stride kernel, no padding).
MaxPool2d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.MaxPool2d
Pools kernel x kernel windows with the given stride and padding.
maxPool3d(Tensor, int, int, int) - Static method in class ai.beestream.jitnn.F
Max pooling of x [N, C, D, H, W] over cubic windows of kernel (padding never wins), PyTorch's max_pool3d.
MaxPool3d - Class in ai.beestream.jitnn.nn
Max pooling over 3-D windows, PyTorch's nn.MaxPool3d.
MaxPool3d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.MaxPool3d
Pools cubic windows of kernel with the given stride and padding.
mean() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
mean() - Method in class ai.beestream.jitnn.Tensor
The mean of all elements, as a scalar.
mean(int[], boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
mean(int, boolean) - Method in class ai.beestream.jitnn.Tensor
The mean along axis, which is kept with size 1 when keepDim is set.
MEAN - Enum constant in enum class ai.beestream.jitnn.F.BagMode
The mean of the rows.
MEAN - Enum constant in enum class ai.beestream.jitnn.F.Reduction
The mean of the per-element losses.
message() - Method in class ai.beestream.jitnn.onnx.Proto
The current length-delimited field as a nested message.
metadata() - Method in class ai.beestream.jitnn.gguf.GgufFile
All metadata, in file order.
metadata() - Method in record class ai.beestream.jitnn.io.SafeTensors.Header
Returns the value of the metadata record component.
min() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
min(int[], boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
min(int, boolean) - Method in class ai.beestream.jitnn.Tensor
The minimum along axis, which is kept with size 1 when keepDim is set; Tensor.argmin(int) gives its index.
MIN - Enum constant in enum class ai.beestream.jitnn.dist.Collective.ReduceOp
The minimum over ranks.
minimum(Tensor) - Method in class ai.beestream.jitnn.Tensor
The element-wise minimum of this and o, with broadcasting (Tensor.minimumNaN(Tensor) propagates NaN).
minimum(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
minimum(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
minimumNaN(Tensor) - Method in class ai.beestream.jitnn.Tensor
Element-wise minimum propagating NaN (torch.minimum).
mish() - Method in class ai.beestream.jitnn.Tensor
Mish: x tanh(softplus(x)).
Mish - Class in ai.beestream.jitnn.nn
Mish: x tanh(softplus(x)).
Mish() - Constructor for class ai.beestream.jitnn.nn.Mish
 
mlmLogits(Tensor, Tensor, Tensor) - Method in class ai.beestream.jitnn.nn.models.Bert
MLM logits [M, vocab] at the INT32 flat positions rows [M] (indices into B * T): the hidden rows are gathered first, so the head runs on the masked positions only.
mm(Tensor) - Method in record class ai.beestream.jitnn.SparseMatrix.Valued
This matrix times dense [cols, feat].
mm(Tensor, Tensor) - Method in class ai.beestream.jitnn.SparseMatrix
A dense with A's stored values values [nnz] (coordinate order); dense [cols, feat] gives [rows, feat].
Mnist - Class in ai.beestream.jitnn.data
Reads the MNIST IDX files (train-images-idx3-ubyte[.gz] and friends) from a directory.
mobilenetV2(int, int, boolean, float, long) - Static method in class ai.beestream.jitnn.nn.models.MobileNetV2
MobileNetV2 over images of inChannels channels with classes outputs and dropout dropout before the classifier, initialised from seed.
MobileNetV2 - Class in ai.beestream.jitnn.nn.models
MobileNetV2 (Sandler et al., 2018), torchvision's layout: inverted residual blocks (1x1 expand, 3x3 depthwise, 1x1 linear projection) with ReLU6.
mod(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
mod(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
model() - Method in class ai.beestream.jitnn.onnx.OnnxSession
The model this session runs.
MODELS - Enum constant in enum class ai.beestream.jitnn.JitnnConfig.Section
The ONNX, TensorFlow and DJL modules.
modi(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
modi(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
module() - Method in class ai.beestream.jitnn.nn.DataParallel
The wrapped replica.
Module - Class in ai.beestream.jitnn.nn
A differentiable building block with parameters and child modules.
Module() - Constructor for class ai.beestream.jitnn.nn.Module
 
mseLoss(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Mean squared error between input and target (same shape).
mseLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Squared error between input and target (same shape), reduced as reduction says.
msleLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Mean squared logarithmic error: (log1p(input) - log1p(target))^2 (Keras's MSLE).
mu() - Method in record class ai.beestream.jitnn.nn.models.Vae.Output
Returns the value of the mu record component.
mul(float) - Method in class ai.beestream.jitnn.Tensor
this * c.
mul(Tensor) - Method in class ai.beestream.jitnn.Tensor
this * o, element-wise with broadcasting.
mul(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
mul(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
muli(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
muli(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
multiHeadAttention(Tensor, Tensor, Tensor, Tensor, Tensor, int, Tensor, boolean) - Static method in class ai.beestream.jitnn.F
Multi-head self-attention on x [B, T, D], as PyTorch's nn.MultiheadAttention with batch_first: q, k, v = split(x Wqkv^T + bqkv), heads heads, then the output projection.
MultiheadAttention - Class in ai.beestream.jitnn.nn
Multi-head self-attention on [B, T, D] inputs, like PyTorch's nn.MultiheadAttention with batch_first: one fused q/k/v projection [3D, D], heads heads of D / heads features, and an output projection.
MultiheadAttention(int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.MultiheadAttention
Attention over dim features in heads heads (dim divisible by it), with or without projection biases, initialised from seed.
multinomial(Tensor, Tensor, int) - Static method in class ai.beestream.jitnn.F
One INT32 index per row of non-negative weights [..., V] (probabilities, or any unnormalised weights), drawn in proportion to the weight.
MULTINOMIAL - Enum constant in enum class ai.beestream.jitnn.ml.NaiveBayes.Kind
A multinomial distribution of counts per class (MultinomialNB).
multiplesOf(int, int) - Static method in class ai.beestream.jitnn.Buckets
The multiples of step up to and including max (rounded up to a multiple).
multiStep(Optimizer, int[], float) - Static method in class ai.beestream.jitnn.nn.optim.LRScheduler
lr = base * gamma^m where m counts the milestones passed.

N

NAdam - Class in ai.beestream.jitnn.nn.optim
NAdam (Dozat), PyTorch's semantics: momentum decay psi (default 0.004), L2 weight decay.
NAdam(List, float, float, float, float, float, float) - Constructor for class ai.beestream.jitnn.nn.optim.NAdam
NAdam with the given betas, eps, L2 weight decay and momentum decay psi.
NaiveBayes - Class in ai.beestream.jitnn.ml
Naive Bayes classifiers after scikit-learn's GaussianNB (per-class means and variances, the variances smoothed by varSmoothing times the largest feature variance) and MultinomialNB (feature counts with additive smoothing alpha).
NaiveBayes(NaiveBayes.Kind, float) - Constructor for class ai.beestream.jitnn.ml.NaiveBayes
smoothing: varSmoothing (Gaussian, scikit-learn's 1e-9) or alpha (multinomial, 1).
NaiveBayes.Kind - Enum Class in ai.beestream.jitnn.ml
Which likelihood the features follow.
name() - Method in class ai.beestream.jitnn.CustomOp
The operator's name, unique among all operators.
name() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns the value of the name record component.
name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the name record component.
name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns the value of the name record component.
name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns the value of the name record component.
name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Returns the value of the name record component.
name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Returns the value of the name record component.
name(DType) - Static method in class ai.beestream.jitnn.io.SafeTensors
The safetensors name of dtype; GGUF block formats have none.
nanToNum(float, float) - Method in class ai.beestream.jitnn.Tensor
NaN to nan, +inf to posInf and -inf to -posInf (torch.nan_to_num with a symmetric infinity).
narrow(int, int, int) - Method in class ai.beestream.jitnn.Tensor
Elements start .. start + length - 1 along axis, as a new tensor.
NATIVE - Enum constant in enum class ai.beestream.jitnn.JitnnConfig.Section
Vendor libraries (cuBLAS, cuDNN, CUTLASS, MLX and others).
NATIVE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.
NATIVE_APPROXIMATE_GELU - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.approximateGelu.
NATIVE_ATTENTION_FP16 - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.attentionFp16.
NATIVE_BIAS_EPILOGUE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.biasEpilogue.
NATIVE_CONV_MIN_FLOPS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.convMinFlops.
NATIVE_CUDNN_CHANNELS_LAST - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.cudnnChannelsLast.
NATIVE_MLX_CONV_MIN_FLOPS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.mlxConvMinFlops.
NATIVE_MLX_COPY_MIN_FLOPS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.mlxCopyMinFlops.
NATIVE_MLX_GATHER_MIN - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.mlxGatherMin.
NATIVE_TF32 - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.tf32.
NATIVE_UNAVAILABLE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.unavailable.
NATIVE_VERBOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.native.verbose.
NcclCollective - Class in ai.beestream.jitnn.dist
Collectives on the GPUs, through NCCL (TornadoVM's tornado-nccl): one rank per CUDA device, in one process.
NcclCollective(String...) - Constructor for class ai.beestream.jitnn.dist.NcclCollective
A communicator over devices ("driver:device" as for tornado --devices), rank r on devices[r]: distinct CUDA GPUs.
NcclCollective(String, int, int, String...) - Constructor for class ai.beestream.jitnn.dist.NcclCollective
This process's share of a communicator over several processes (on one machine or many): ranks firstRank, firstRank + 1, ... on devices, out of worldSize.
ne(Tensor) - Method in class ai.beestream.jitnn.Tensor
1 where this !
nearest(float[], int, int) - Static method in class ai.beestream.jitnn.nn.text.Corpus
The row of vectors [n, dim] most similar to row of (by cosine), excluding itself.
neg() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
neg() - Method in class ai.beestream.jitnn.Tensor
-this.
negatives(int) - Method in class ai.beestream.jitnn.nn.text.Corpus
n words drawn from the unigram distribution to the power 3/4.
negativeSamplingLoss(Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Negative-sampling loss of word2vec (Mikolov et al., 2013): hidden [B, D] (the centre word, or the averaged context), target [B, D] (the output vector of the true word) and negatives [B, K, D] (output vectors of sampled words); mean over B of -log sigmoid(h . t) - sum_k log sigmoid(-h . n_k).
negi() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
neq(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
neq(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
newBaseManager() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
newBaseManager(Device) - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
newGradientCollector() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
newModel(String, Device) - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
newSubManager(Device) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
newSymbolBlock(NDManager) - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
newTrainer(TrainingConfig) - Method in class ai.beestream.jitnn.djl.JitnnModel
 
next() - Method in class ai.beestream.jitnn.onnx.Proto
Advances to the next field; false at the end of the message.
nll(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Per-sample negative log-likelihood of INT32 labels under log-probabilities logp.
nllLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Negative log-likelihood of INT32 labels under log-probabilities [N, C], reduced.
nnz() - Method in class ai.beestream.jitnn.SparseMatrix
The number of stored entries.
node() - Method in class ai.beestream.jitnn.engine.Emitter
The node whose tasks are being emitted, or null.
Node - Class in ai.beestream.jitnn.engine
One operator application.
Node(Op, Tensor[], Tensor[], int[], float[], boolean) - Constructor for class ai.beestream.jitnn.engine.Node
An application of op; a null attribute array becomes an empty one.
Node(String, String, String, List, List, Map) - Constructor for record class ai.beestream.jitnn.onnx.OnnxModel.Node
Creates an instance of a Node record class.
nodes() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns the value of the nodes record component.
NONE - Enum constant in enum class ai.beestream.jitnn.F.Reduction
The per-element losses, unreduced.
nonzero() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
nonzero() - Method in class ai.beestream.jitnn.Tensor
INT32 coordinates [count, rank] of the non-zero elements (torch.nonzero); reads the count back to the host.
nonzero(Tensor, int, int) - Static method in class ai.beestream.jitnn.F
torch.nonzero_static: INT32 coordinates [size, rank] of the first size non-zero elements, the rest of the rows fill.
norm(boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
norm(int, int[], boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
norm(int, int, boolean) - Method in class ai.beestream.jitnn.Tensor
The p-norm (p = 1 or 2) along axis: sum |x| or sqrt(sum x^2).
normal(long, float, float, int...) - Static method in class ai.beestream.jitnn.Tensor
Normal with the given mean and standard deviation, counter-based like Tensor.uniform(long, float, float, int...).
normal(Tensor, float, float, int...) - Static method in class ai.beestream.jitnn.Tensor
As Tensor.normal(long, float, float, int...) with the seed in an INT32 [2] tensor.
normalize(double, long, double) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
normalize(float[], float[]) - Static method in class ai.beestream.jitnn.data.Transforms
(x - mean[c]) / std[c] per channel (torchvision's Normalize).
normalizeImage(Tensor, float[], float[]) - Static method in class ai.beestream.jitnn.F
Normalises images x [N, C, H, W] per channel: (x - mean[c]) / std[c], as torchvision's Normalize.
numel() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Number of values.
numel() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
The number of elements: the product of the shape.
numel() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
The number of elements: the product of the dims.
numel() - Method in class ai.beestream.jitnn.Shape
The number of elements: the product of the sizes (1 for a scalar).
numel() - Method in class ai.beestream.jitnn.Tensor
The number of elements (1 for a scalar).

O

of(float[], int...) - Static method in class ai.beestream.jitnn.Tensor
An FP32 tensor with a copy of data in the shape dims (a vector of all the values when no dims are given).
of(float[], int[], int...) - Static method in interface ai.beestream.jitnn.data.Dataset
A dataset over arrays: x holds the samples back to back, y one label each.
of(int...) - Static method in class ai.beestream.jitnn.Buckets
These sizes.
of(int...) - Static method in class ai.beestream.jitnn.Shape
The shape with the given sizes (the array is copied).
of(int[], int...) - Static method in class ai.beestream.jitnn.Tensor
An INT32 tensor (labels, indices) with a copy of data; Tensor.of(float[], int...) for int data.
of(OnnxModel.Graph, Map) - Static method in class ai.beestream.jitnn.onnx.OnnxModel
A model from a graph built in memory (by jitNN's other importers), with its operator-set versions.
of(OnnxModel, boolean) - Static method in class ai.beestream.jitnn.onnx.OnnxSession
A session over a model built in memory (see OnnxModel.of(OnnxModel.Graph, Map)); halfPrecision as in OnnxSession.load(Path, boolean).
of(Tensor[], int[]) - Method in interface ai.beestream.jitnn.CustomOp.OutputShape
The output's shape for these inputs and integer attributes.
of(HalfFloatArray, int...) - Static method in class ai.beestream.jitnn.Tensor
An FP16 input over data, held without a copy (an FP16 image batch for half-precision inference, say): plans read it as they read any host input.
offset() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns the value of the offset record component.
offset() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns the value of the offset record component.
onCuda() - Static method in class ai.beestream.jitnn.JitnnRuntime
True when this thread's plans run through TornadoVM on a CUDA device.
onDevice(String, Supplier) - Static method in class ai.beestream.jitnn.JitnnRuntime
Runs body with this thread's plans on device spec ("driver:device", as jitnn.device), syncs, then restores the previous device.
oneHot(int, float, float, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
oneHot(Tensor, int) - Static method in class ai.beestream.jitnn.F
INT32 labels [...] to a one-hot [..., classes].
ones(int...) - Static method in class ai.beestream.jitnn.Tensor
An FP32 tensor of ones.
ones(Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
onGradReady(Consumer) - Method in class ai.beestream.jitnn.Tensor
Calls hook with this leaf tensor (a parameter) during each backward pass, as soon as its gradient is final: after the last operator that contributes to it, while the rest of the backward pass is still being recorded.
ONNX_BUCKETS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.onnx.buckets.
ONNX_FUSED_ATTENTION_ROWS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.onnx.fusedAttentionRows.
ONNX_HALF_ACTIVATIONS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.onnx.halfActivations.
ONNX_INT8 - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.onnx.int8.
ONNX_REWRITE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.onnx.rewrite.
OnnxExport - Class in ai.beestream.jitnn.onnx
Writes a jitNN forward pass as an ONNX model (opset 20), for ONNX Runtime and the other ONNX tools.
OnnxModel - Class in ai.beestream.jitnn.onnx
An ONNX model as read from its file: the main graph's nodes, initializers, inputs and outputs, and the operator-set versions.
OnnxModel.Attribute - Record Class in ai.beestream.jitnn.onnx
An attribute of a node; which fields are set depends on its type.
OnnxModel.Graph - Record Class in ai.beestream.jitnn.onnx
A graph: its nodes in topological order, the initializers (weights and constants) by name, and its inputs and outputs.
OnnxModel.Node - Record Class in ai.beestream.jitnn.onnx
A node of the graph: an operator application, with the names of its input and output values.
OnnxModel.TensorData - Record Class in ai.beestream.jitnn.onnx
A tensor stored in the model: dims, element type and its data as little-endian bytes.
OnnxModel.ValueInfo - Record Class in ai.beestream.jitnn.onnx
A graph input or output: name, element type and dims (-1 for a symbolic dimension), with the symbolic dimensions' names (dim_param, null where a dimension has none or dims is null).
OnnxSession - Class in ai.beestream.jitnn.onnx
Runs an ONNX model with jitNN: each node becomes jitNN operations, so a call records one trace that jitNN compiles into a TornadoVM plan on the first run and replays afterwards (same input shapes).
op - Variable in class ai.beestream.jitnn.engine.Node
The operator applied.
open(Path) - Static method in class ai.beestream.jitnn.gguf.GgufFile
Reads the metadata and the tensor list of path; tensors are mapped when loaded.
open(Path) - Static method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Opens a model directory.
opset(String) - Method in class ai.beestream.jitnn.onnx.OnnxModel
The operator-set version of domain ("" is the default ONNX domain).
optimizer - Variable in class ai.beestream.jitnn.nn.optim.LRScheduler
The optimiser whose rate this schedule sets.
Optimizer - Class in ai.beestream.jitnn.nn.optim
Updates parameters in place from their gradients.
Optimizer(List, float) - Constructor for class ai.beestream.jitnn.nn.optim.Optimizer
An optimiser over params with learning rate lr.
options() - Static method in class ai.beestream.jitnn.JitnnConfig
Every option, in documentation order.
opType() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns the value of the opType record component.
originalContext() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Returns the value of the originalContext record component.
ORTHO - Enum constant in enum class ai.beestream.jitnn.Fft.Norm
Both transforms scaled by 1/sqrt(n).
out(int) - Method in class ai.beestream.jitnn.engine.Node
Output i.
outProjection() - Method in class ai.beestream.jitnn.nn.MultiheadAttention
The output projection [dim, dim].
output(CustomOp.OutputShape, DType) - Method in class ai.beestream.jitnn.CustomOp.Builder
Adds an output of the given shape and element type.
Output(Tensor, Tensor) - Constructor for record class ai.beestream.jitnn.nn.GRU.Output
Creates an instance of a Output record class.
Output(Tensor, Tensor, Tensor) - Constructor for record class ai.beestream.jitnn.nn.LSTM.Output
Creates an instance of a Output record class.
Output(Tensor, Tensor, Tensor) - Constructor for record class ai.beestream.jitnn.nn.models.Vae.Output
Creates an instance of a Output record class.
outputNames() - Method in class ai.beestream.jitnn.onnx.OnnxSession
The names of the graph's outputs, in graph order.
outputNames() - Method in class ai.beestream.jitnn.tf.TfModel
The model's output names.
outputs - Variable in class ai.beestream.jitnn.engine.Node
The operator's outputs.
outputs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns the value of the outputs record component.
outputs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns the value of the outputs record component.
overlapGradients(long) - Method in class ai.beestream.jitnn.nn.DataParallel
Overlaps the gradient all-reduce with the backward pass, as PyTorch's DistributedDataParallel: the parameters are grouped, in reverse order (the order their gradients become final), into buckets of about bucketBytes; a bucket's all-reduce starts as soon as its last gradient is final, while the backward pass of the earlier layers goes on, and DataParallel.averageGradients() only waits for it.

P

pad(Tensor, int[], int[], F.PadMode, float) - Static method in class ai.beestream.jitnn.F
Pads the spatial axes of x [N, C, ...]: before[k] and after[k] elements on spatial axis k, PyTorch's F.pad for constant, reflect and replicate modes.
pad(Tensor, int, int, float) - Static method in class ai.beestream.jitnn.Buckets
t with dimension axis grown to size, the new elements set to value; built on the host (a device result is read back first).
pad(Shape, double) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
Pad - Class in ai.beestream.jitnn.nn
Padding of the spatial axes: constant, reflect or replicate (PyTorch's nn.ConstantPadNd, ReflectionPadNd, ReplicationPadNd).
Pad(int[], int[], F.PadMode, float) - Constructor for class ai.beestream.jitnn.nn.Pad
Pads spatial axis k with before[k] and after[k] elements; value fills in constant mode.
paddedDimensions() - Method in class ai.beestream.jitnn.onnx.OnnxSession
The symbolic dimensions OnnxSession.withBuckets(int...) pads (for tests and logging).
padId() - Method in class ai.beestream.jitnn.nn.text.Tokenizer
The id used for padding: the file's padding id, else [PAD] or <pad>, else GPT-2's <|endoftext|>, else 0.
pairwiseDistance(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Euclidean distance of each row pair (last axis), PyTorch's pairwise_distance with eps 1e-6.
PARALLEL_COPY_BYTES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.parallelCopyBytes.
parameter(Tensor) - Method in class ai.beestream.jitnn.nn.Module
Registers a trainable parameter initialised from init.
parameter(Tensor) - Static method in class ai.beestream.jitnn.Tensor
A trainable parameter: persistent storage (it stays on the device across steps), gradient tracking on, initialised with the values of init.
parameterCount() - Method in class ai.beestream.jitnn.nn.Module
The number of trainable values: the elements of every parameter.
parameters() - Method in class ai.beestream.jitnn.nn.Module
This module's parameters and its children's, in registration order.
params - Variable in class ai.beestream.jitnn.nn.optim.Optimizer
The parameters this optimiser updates.
parse(String) - Static method in class ai.beestream.jitnn.io.Json
Parses one JSON value.
parse(String) - Method in class ai.beestream.jitnn.JitnnConfig.Option
Parses a property value, failing with a message that names the key.
parse(Path) - Static method in class ai.beestream.jitnn.io.Json
Parses a UTF-8 JSON file.
path() - Method in class ai.beestream.jitnn.gguf.GgufFile
The file this was read from.
Pca - Class in ai.beestream.jitnn.ml
Principal component analysis from the eigendecomposition of the covariance matrix, after scikit-learn's PCA: components as rows, ordered by explained variance, each signed so that its largest-magnitude entry is positive (scikit-learn's svd_flip convention).
Pca(int) - Constructor for class ai.beestream.jitnn.ml.Pca
PCA keeping components components.
pending() - Static method in class ai.beestream.jitnn.JitnnRuntime
Number of operators recorded and not yet run on this thread.
permute(int...) - Method in class ai.beestream.jitnn.Tensor
The axes reordered: axis i of the result is axis order[i] of this tensor.
persistent(Tensor) - Static method in class ai.beestream.jitnn.Tensor
Persistent storage without gradient tracking (optimiser state, running statistics).
persistent(TornadoNativeArray, DType, int...) - Static method in class ai.beestream.jitnn.Tensor
A persistent tensor (a weight that stays on the device) over an existing native array, without copying it: for example an array that wraps a memory-mapped file.
persistentHalf(float[], int...) - Static method in class ai.beestream.jitnn.Tensor
Persistent FP16 storage (weights kept in half precision on the device), from FP32 values rounded to nearest.
PIXELS - Static variable in class ai.beestream.jitnn.data.Cifar10
Values per image: 3 channels of 32 x 32.
PIXELS - Static variable in class ai.beestream.jitnn.data.Mnist
Pixels per image.
PLAN_CACHE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.plan.cache.
PLAN_CARRY - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.plan.carry.
PLAN_IDLE_BYTES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.plan.idleBytes.
PLAN_IDLE_EXECUTIONS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.plan.idleExecutions.
PLAN_MEMO - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.plan.memo.
planStats() - Static method in class ai.beestream.jitnn.JitnnRuntime
Plans compiled, replayed and reused for released storages on this JVM so far, and the replays that skipped the rewrite: [compiles, replays, reuses, remembered].
poissonNllLoss(Tensor, Tensor, boolean, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Poisson negative log-likelihood, PyTorch's poisson_nll_loss without the Stirling term: exp(input) - target input with logInput, else input - target log(input + eps).
pool(Tensor, F.PoolMode, int[], int[], int[], int[], boolean, boolean, float) - Static method in class ai.beestream.jitnn.F
Pooling over 1 to 3 spatial axes of x [N, C, ...] with a size per axis, as ONNX's MaxPool, AveragePool and LpPool and PyTorch's pooling layers: pads holds the padding before every axis, then after every axis; ceil rounds the output size up (a window must still start inside the input or its leading padding); average pooling counts the padding when countIncludePad (never the positions added by ceil); LP pooling computes (sum |x|^p)^(1/p).
pooled(Tensor) - Method in class ai.beestream.jitnn.nn.models.Bert
The pooler output [B, dim]: tanh(dense(hidden state of the first token)); needs the pooler.
pow(float) - Method in class ai.beestream.jitnn.Tensor
x^exponent, element-wise.
pow(Tensor) - Method in class ai.beestream.jitnn.Tensor
x^exponent, element-wise, with broadcasting.
pow(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
pow(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
powersOfTwo(int, int) - Static method in class ai.beestream.jitnn.Buckets
The powers of two from min up to and including max (both rounded up to a power of two).
powi(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
powi(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
predict(Tensor) - Method in class ai.beestream.jitnn.ml.GradientBoosting
The most probable label of each sample.
predict(Tensor) - Method in class ai.beestream.jitnn.ml.KMeans
Index of the nearest centre of each row.
predict(Tensor) - Method in class ai.beestream.jitnn.ml.KNeighbors
The majority label of each query's neighbours (a classifier's fit only).
predict(Tensor) - Method in class ai.beestream.jitnn.ml.LinearRegression
Predictions x w + b for samples x [n, features].
predict(Tensor) - Method in class ai.beestream.jitnn.ml.LinearSvc
The predicted label of each sample.
predict(Tensor) - Method in class ai.beestream.jitnn.ml.LogisticRegression
The most probable label of each sample.
predict(Tensor) - Method in class ai.beestream.jitnn.ml.NaiveBayes
The most probable label of each sample.
predict(Tensor) - Method in class ai.beestream.jitnn.ml.RandomForest
The label with the highest averaged probability, for each sample.
predictProba(Tensor) - Method in class ai.beestream.jitnn.ml.GradientBoosting
Class probabilities [n, classes] for samples x.
predictProba(Tensor) - Method in class ai.beestream.jitnn.ml.KNeighbors
Class probabilities [m, c]: the fraction of neighbours per class.
predictProba(Tensor) - Method in class ai.beestream.jitnn.ml.LogisticRegression
Class probabilities [m, c].
predictProba(Tensor) - Method in class ai.beestream.jitnn.ml.NaiveBayes
Class probabilities [n, classes] for samples x.
predictValues(Tensor) - Method in class ai.beestream.jitnn.ml.GradientBoosting
Regression predictions [m].
predictValues(Tensor) - Method in class ai.beestream.jitnn.ml.KNeighbors
Regression predictions [m, t].
predictValues(Tensor) - Method in class ai.beestream.jitnn.ml.RandomForest
Mean of the trees' leaf values: class probabilities [m, c] or predictions [m, 1].
prefetch(int) - Method in class ai.beestream.jitnn.data.DataLoader.Builder
Batches assembled ahead of the one being used, with workers (default 2).
prelu(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
PReLU: x where x > 0, else a x, with a learnable slope weight of one element or one per channel (axis 1 of x, as PyTorch).
PReLU - Class in ai.beestream.jitnn.nn
PReLU with a learnable slope, one shared or one per channel, initialised to init (PyTorch: 0.25).
PReLU(int, float) - Constructor for class ai.beestream.jitnn.nn.PReLU
PReLU with parameters slopes (1 shared, or one per channel), each init at the start.
PRINT_KERNELS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.printKernels.
printKernels(PrintStream) - Static method in class ai.beestream.jitnn.JitnnRuntime
Prints every plan compiled from now on to out (null stops): the plan's operators with their shapes and task counts, then the CUDA, OpenCL or Metal source of each kernel.
ProcessGroup - Class in ai.beestream.jitnn.dist
Runs one rank per device in this process: rank firstRank + r is a thread whose plans run on devices[r] (see Engine.onDevice(String, Supplier)), and all ranks share one Collective.
prod() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
prod() - Method in class ai.beestream.jitnn.Tensor
Product of all elements.
prod(int[], boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
prod(int, boolean) - Method in class ai.beestream.jitnn.Tensor
Product along axis.
PROFILE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.profile.
PROMOTE_INPUT_BYTES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.promoteInputBytes.
Proto - Class in ai.beestream.jitnn.onnx
A reader for the protocol buffers wire format, enough to decode ONNX's messages without a protobuf library: varints, fixed 32/64-bit values and length-delimited fields (strings, bytes, nested messages and packed repeated numbers).
Proto(ByteBuffer) - Constructor for class ai.beestream.jitnn.onnx.Proto
A reader over the message in buf, from its position to its limit.
Proto.DoubleSink - Interface in ai.beestream.jitnn.onnx
Receives double values from Proto.doubles(Proto.DoubleSink).
Proto.FloatSink - Interface in ai.beestream.jitnn.onnx
Receives float values from Proto.floats(Proto.FloatSink).
Proto.LongSink - Interface in ai.beestream.jitnn.onnx
Receives int64 values from Proto.longs(Proto.LongSink).
put(NDArray, NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
put(String, float) - Method in class ai.beestream.jitnn.io.StateDict
Adds a float value (stored as its decimal string).
put(String, long) - Method in class ai.beestream.jitnn.io.StateDict
Adds an integer value (stored as its decimal string).
put(String, StateDict) - Method in class ai.beestream.jitnn.io.StateDict
Adds every entry of part under prefix..
put(String, Tensor) - Method in class ai.beestream.jitnn.io.StateDict
Adds a tensor; its contents are read when the dictionary is saved.
put(String, String) - Method in class ai.beestream.jitnn.io.StateDict
Adds a value.
PV_DBOW - Enum constant in enum class ai.beestream.jitnn.nn.models.Doc2Vec.Mode
From the document vector alone (distributed bag of words).
PV_DM - Enum constant in enum class ai.beestream.jitnn.nn.models.Doc2Vec.Mode
From the mean of the document vector and the context words' vectors (distributed memory).

Q

Q2_K - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 2-bit k-quant super-blocks of 256 values (84 bytes).
Q2_K - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
2-bit k-quant super-blocks of 256 values.
Q3_K - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 3-bit k-quant super-blocks of 256 values (110 bytes).
Q3_K - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
3-bit k-quant super-blocks of 256 values.
Q4_0 - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 4-bit blocks of 32 values with an FP16 scale (18 bytes).
Q4_0 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
4-bit blocks of 32 values with an FP16 scale.
Q4_1 - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 4-bit blocks of 32 values with an FP16 scale and minimum (20 bytes).
Q4_1 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
4-bit blocks of 32 values with an FP16 scale and minimum.
Q4_K - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 4-bit k-quant super-blocks of 256 values (144 bytes).
Q4_K - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
4-bit k-quant super-blocks of 256 values.
Q5_0 - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 5-bit blocks of 32 values with an FP16 scale (22 bytes).
Q5_0 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
5-bit blocks of 32 values with an FP16 scale.
Q5_1 - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 5-bit blocks of 32 values with an FP16 scale and minimum (24 bytes).
Q5_1 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
5-bit blocks of 32 values with an FP16 scale and minimum.
Q5_K - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 5-bit k-quant super-blocks of 256 values (176 bytes).
Q5_K - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
5-bit k-quant super-blocks of 256 values.
Q6_K - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 6-bit k-quant super-blocks of 256 values (210 bytes).
Q6_K - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
6-bit k-quant super-blocks of 256 values.
q8(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
Q8_0 tensors: GGUF's 34-byte blocks of 32 values.
Q8_0 - Enum constant in enum class ai.beestream.jitnn.DType
GGUF's 8-bit blocks: per 32 values an FP16 scale and 32 signed bytes (34 bytes).
Q8_0 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
8-bit blocks of 32 values with an FP16 scale.
Q8_1 - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
8-bit blocks of 32 values with a scale and sum; raw bytes only.
Q8_K - Enum constant in enum class ai.beestream.jitnn.gguf.GgmlType
8-bit k-quant super-blocks of 256 values (llama.cpp's intermediate type); raw bytes only.
qkv() - Method in class ai.beestream.jitnn.nn.MultiheadAttention
The fused q/k/v input projection [3 dim, dim].
qr(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Reduced QR, torch.linalg.qr: Q [..., m, k] and R [..., k, n], k = min(m, n).
quantized(DType, byte[], int...) - Static method in class ai.beestream.jitnn.Tensor
A weight in one of GGUF's block-quantised formats (Q8_0, Q4_0, Q4_1, Q4_K, Q5_K, Q6_K, Q5_0, Q5_1, Q2_K, Q3_K) from its raw blocks, as a GGUF file stores them: persistent, so it stays on the device.
quantizeQ8(Tensor) - Static method in class ai.beestream.jitnn.F
FP32 x (last axis a multiple of 32) quantised to Q8_0, GGUF's 8-bit block format.

R

randomCrop(int) - Static method in class ai.beestream.jitnn.data.Transforms
Pads the image with pad zeros on every side and crops the original size back at a random offset (torchvision's RandomCrop(size, padding)).
RandomForest - Class in ai.beestream.jitnn.ml
Random forests (Breiman, 2001) and single decision trees on binned features: trees grown on bootstrap samples (weights drawn on the host) with a random subset of the features at each split, predictions averaged.
RandomForest(RandomForest.Task, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.ml.RandomForest
maxFeatures <= 0: sqrt(F) for classification, all for regression (scikit-learn's defaults).
RandomForest(RandomForest.Task, int, int, long) - Constructor for class ai.beestream.jitnn.ml.RandomForest
A forest of trees bootstrapped trees of depth at most maxDepth, one sample per leaf at least, scikit-learn's default feature subsets, 128 bins.
RandomForest.Task - Enum Class in ai.beestream.jitnn.ml
What the forest predicts.
randomHorizontalFlip() - Static method in class ai.beestream.jitnn.data.Transforms
Mirrors the image left to right with probability 1/2 (torchvision's RandomHorizontalFlip).
randomInteger(long, long, Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
randomNormal(float, float, Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
randomPermutation(long) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
randomUniform(float, float, Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
rank() - Method in class ai.beestream.jitnn.nn.DataParallel
This replica's rank.
rank() - Method in class ai.beestream.jitnn.Shape
The number of axes.
rank() - Method in class ai.beestream.jitnn.Tensor
The number of axes (0 for a scalar).
rate(int) - Method in class ai.beestream.jitnn.nn.optim.LRScheduler
The learning rate for epoch (0-based).
raw() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Returns the value of the raw record component.
raw(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
The raw bytes of tensor name, mapped read-only and released when unreachable.
read(Path, SafeTensors.Entry, MemorySegment) - Static method in class ai.beestream.jitnn.io.SafeTensors
Reads entry's bytes from file into target, which must be at least that long.
readHeader(Path) - Static method in class ai.beestream.jitnn.io.SafeTensors
Reads the header of file.
reciprocal() - Method in class ai.beestream.jitnn.Tensor
1 / x, element-wise.
recon() - Method in record class ai.beestream.jitnn.nn.models.Vae.Output
Returns the value of the recon record component.
record(Op, Tensor[], Tensor[], int[], float[], boolean) - Static method in class ai.beestream.jitnn.Tensor
Records an operator application.
reduce(int, boolean) - Method in class ai.beestream.jitnn.Shape
The shape with axis removed (keepDim=false) or set to 1 (keepDim=true).
reduceScatter(int, Tensor, Tensor, Collective.ReduceOp) - Method in interface ai.beestream.jitnn.dist.Collective
input [worldSize * n, ...] on every rank; output [n, ...] receives on rank r the reduction of every rank's r-th slice.
reduceScatter(int, Tensor, Tensor, Collective.ReduceOp) - Method in class ai.beestream.jitnn.dist.HostCollective
 
reduceScatter(int, Tensor, Tensor, Collective.ReduceOp) - Method in class ai.beestream.jitnn.dist.NcclCollective
Persistent output [n, ...] receives the reduction of every rank's slice r of input [world * n, ...].
reference(CustomOp.Reference) - Method in class ai.beestream.jitnn.CustomOp.Builder
The host implementation, used by the reference backend and by tests (optional).
REFERENCE - Enum constant in enum class ai.beestream.jitnn.Backend
Operators run eagerly on the host with the Java reference implementations.
REFLECT - Enum constant in enum class ai.beestream.jitnn.F.PadMode
Mirrors the input about its edge, without repeating the edge element.
REGRESSION - Enum constant in enum class ai.beestream.jitnn.ml.RandomForest.Task
Real values, by the averaged tree outputs.
release() - Method in class ai.beestream.jitnn.Tensor
Declares the tensor no longer used; using it, or a view of it, afterwards is an error.
relu() - Method in class ai.beestream.jitnn.Tensor
max(x, 0), element-wise.
ReLU - Class in ai.beestream.jitnn.nn
ReLU: max(x, 0).
ReLU() - Constructor for class ai.beestream.jitnn.nn.ReLU
 
relu6() - Method in class ai.beestream.jitnn.Tensor
min(max(x, 0), 6).
ReLU6 - Class in ai.beestream.jitnn.nn
ReLU6: min(max(x, 0), 6).
ReLU6() - Constructor for class ai.beestream.jitnn.nn.ReLU6
 
remainder(Tensor) - Method in class ai.beestream.jitnn.Tensor
this - floor(this / o) o, with broadcasting: the remainder with the sign of the divisor (torch.remainder).
repeat(int, long) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
repeat(long) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
repeat(long[]) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
REPLICATE - Enum constant in enum class ai.beestream.jitnn.F.PadMode
Repeats the edge element.
requiresGrad() - Method in class ai.beestream.jitnn.Tensor
Whether gradients are tracked for this tensor.
requiresGrad(boolean) - Method in class ai.beestream.jitnn.Tensor
Turns gradient tracking on or off for a leaf tensor.
reservePlans(int) - Static method in class ai.beestream.jitnn.JitnnRuntime
Makes room for at least plans more compiled plans in the plan cache (per concurrency slot), beyond jitnn.plan.cache: for callers that cycle through many shapes, as shape buckets do, so their plans are not evicted and compiled again.
reset() - Static method in class ai.beestream.jitnn.JitnnRuntime
Closes every cached compiled plan, first saving persistent storages (weights) to the host; the next run of each trace compiles again.
reshape(int...) - Method in class ai.beestream.jitnn.Tensor
Same data, new shape; one dimension may be -1.
reshape(Shape) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
ResNet - Class in ai.beestream.jitnn.nn.models
ResNet (He et al., 2016).
resnet18(int, int, boolean, long) - Static method in class ai.beestream.jitnn.nn.models.ResNet
ResNet-18: BasicBlocks [2, 2, 2, 2], widths 64-512.
resnet50(int, int, boolean, long) - Static method in class ai.beestream.jitnn.nn.models.ResNet
ResNet-50: Bottleneck blocks [3, 4, 6, 3], widths 64-512, expansion 4 (torchvision's v1.5, stride on the 3x3).
result(Shape, DType) - Static method in class ai.beestream.jitnn.Tensor
An output tensor for an operator; its storage is allocated when the operator runs.
REUSE_BUFFERS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.reuseBuffers.
REUSE_BUFFERS_EXACT - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.reuseBuffers.exact.
reverseSequences(Tensor, int[]) - Static method in class ai.beestream.jitnn.F
x [T, B, ...] with each sequence reversed within its length (padding left in place), as a bidirectional layer reads it; lengths null reverses the whole time axis.
rfft(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.rfft of real x along the last axis, full length, BACKWARD scaling.
rfft(Tensor, int, int, Fft.Norm) - Static method in class ai.beestream.jitnn.Fft
torch.fft.rfft of real x: the n / 2 + 1 non-negative frequencies, complex [..., n / 2 + 1, 2].
rfft2(Tensor) - Static method in class ai.beestream.jitnn.Fft
torch.fft.rfft2: rfft over the last axis, then fft over the one before.
rgbToGrayscale(Tensor) - Static method in class ai.beestream.jitnn.F
RGB images [N, 3, H, W] to luminance [N, 1, H, W] with torchvision's weights (0.2989, 0.587, 0.114).
rmsNorm(Tensor, Tensor, float) - Static method in class ai.beestream.jitnn.F
RMSNorm over the last axis: x / sqrt(mean(x^2) + eps) * weight.
RMSNorm - Class in ai.beestream.jitnn.nn
RMSNorm over the last axis, with weight 1 initially, like PyTorch's nn.RMSNorm.
RMSNorm(int, float) - Constructor for class ai.beestream.jitnn.nn.RMSNorm
RMSNorm over features values with the given eps.
RMSProp - Class in ai.beestream.jitnn.nn.optim
RMSProp (Hinton), PyTorch's semantics: smoothing alpha, optional momentum and centring, L2 weight decay.
RMSProp(List, float, float, float, float, float, boolean) - Constructor for class ai.beestream.jitnn.nn.optim.RMSProp
RMSProp with smoothing alpha, eps, L2 weight decay, momentum (0 for none) and, with centered, the variance estimated around the mean gradient.
RNN - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.rnn.
rope(Tensor, float, int, boolean) - Static method in class ai.beestream.jitnn.F
Rotary position embedding of x [..., T, D] at positions offset .. offset + T - 1, frequency base^(-2i / D) for pair i.
ropeHeads(Tensor, float, int, boolean) - Static method in class ai.beestream.jitnn.F
Rotary position embedding of x [..., T, H, D] (the heads of a token share its position); see F.rope(Tensor, float, int, boolean).
ropeHeads(Tensor, float, RopeScaling, int, boolean) - Static method in class ai.beestream.jitnn.F
F.ropeHeads(Tensor, float, int, boolean) with the frequencies rescaled for longer contexts (Llama 3.1's rope_scaling; null for none).
ropeScaling(RopeScaling) - Method in class ai.beestream.jitnn.nn.models.Llama
Rescales the RoPE frequencies of every block (Llama 3.1's rope_scaling); null for none.
RopeScaling - Interface in ai.beestream.jitnn
A rescaling of the rotary position embedding's frequencies, as Hugging Face's rope_scaling.
RopeScaling.Linear - Record Class in ai.beestream.jitnn
Positions divided by factor.
RopeScaling.Llama3 - Record Class in ai.beestream.jitnn
round() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
round() - Method in class ai.beestream.jitnn.Tensor
Element-wise rounding to the nearest integer, halves to even (PyTorch's round).
ROWNORM - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.rownorm.
rows() - Method in class ai.beestream.jitnn.SparseMatrix
The number of rows.
rows(Buckets, Tensor, UnaryOperator) - Static method in class ai.beestream.jitnn.nn.Bucketing
model over x [N, ...] with N padded to a batch size by zero rows; the result's first N rows.
rrelu(Tensor, float, float, boolean, Tensor) - Static method in class ai.beestream.jitnn.F
RReLU (Xu et al., 2015): in training, a slope drawn uniformly from [lower, upper) for every element, seeded by the INT32 [2] seed tensor (see Tensor.seed(long)); in evaluation the mean slope (lower + upper) / 2.
RReLU - Class in ai.beestream.jitnn.nn
RReLU: random slopes in [lower, upper) while training (a fresh seed per call, so a replayed step draws new slopes without recompiling), the mean slope in evaluation.
RReLU(float, float, long) - Constructor for class ai.beestream.jitnn.nn.RReLU
RReLU with slopes in [lower, upper), drawn from a stream seeded by seed.
run(Node) - Method in interface ai.beestream.jitnn.CustomOp.Reference
Computes the outputs of node from its inputs, on the host.
run(Tensor, Tensor) - Method in class ai.beestream.jitnn.nn.models.Vae
Encodes x [B, in], samples z = mu + exp(logvar / 2) eps with the given noise eps [B, latent], decodes.
run(String[], Collective, BiFunction) - Static method in class ai.beestream.jitnn.dist.ProcessGroup
Runs body(rank, collective) on every rank and returns the ranks' results in rank order.
run(String[], BiFunction) - Static method in class ai.beestream.jitnn.dist.ProcessGroup
run(Map) - Method in class ai.beestream.jitnn.onnx.OnnxSession
Runs the graph on inputs (by graph input name) and returns every graph output, as lazy jitNN tensors (read them with toFloatArray, or keep computing with them).
run(Map) - Method in class ai.beestream.jitnn.tf.TfModel
Runs the model; outputs are lazy jitNN tensors, by signature output name.
runningMean() - Method in class ai.beestream.jitnn.nn.BatchNorm2d
The running mean [channels], updated in training mode.
runningMean() - Method in class ai.beestream.jitnn.nn.SyncBatchNorm2d
The running mean [channels], updated in training mode.
runningVar() - Method in class ai.beestream.jitnn.nn.BatchNorm2d
The running variance [channels], updated in training mode.
runningVar() - Method in class ai.beestream.jitnn.nn.SyncBatchNorm2d
The running variance [channels], updated in training mode.
RUNTIME - Enum constant in enum class ai.beestream.jitnn.JitnnConfig.Section
Execution, devices, plans and memory.

S

s() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the s record component.
SafeTensors - Class in ai.beestream.jitnn.io
Reads and writes the safetensors format (Hugging Face): an 8-byte little-endian header length, a JSON header that maps each tensor name to its dtype, shape and byte range, optional __metadata__ string pairs, then the raw little-endian data. jitNN's checkpoints use it, so they open in Python with safetensors.torch.load_file.
SafeTensors.Entry - Record Class in ai.beestream.jitnn.io
One tensor of a file: its type (null when jitNN has no such type, e.g.
SafeTensors.Header - Record Class in ai.beestream.jitnn.io
A file's header: the tensors, in file order, and the metadata.
SafeTensors.Source - Record Class in ai.beestream.jitnn.io
The data to write for one tensor: its type, shape, and a segment of exactly its bytes.
sample(int, float[], int, int[], int) - Method in class ai.beestream.jitnn.data.Cifar10
 
sample(int, float[], int, int[], int) - Method in interface ai.beestream.jitnn.data.Dataset
Copies sample i into x at xAt and its label into y at yAt (DataLoader assembles shuffled batches with it).
sample(int, float[], int, int[], int) - Method in class ai.beestream.jitnn.data.Mnist
 
sampleShape() - Method in class ai.beestream.jitnn.data.Cifar10
 
sampleShape() - Method in interface ai.beestream.jitnn.data.Dataset
The shape of one sample, e.g. {1, 28, 28} or {3, 32, 32}.
sampleShape() - Method in class ai.beestream.jitnn.data.Mnist
 
sampleSize() - Method in interface ai.beestream.jitnn.data.Dataset
The number of values in one sample: the product of Dataset.sampleShape().
save(Path) - Method in class ai.beestream.jitnn.io.StateDict
Writes the dictionary to path as safetensors; reading live tensors waits for them.
saveScalars(StateDict) - Method in class ai.beestream.jitnn.nn.optim.NAdam
 
saveScalars(StateDict) - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Scalar state beyond the step count and learning rate, for checkpoints.
scalar() - Static method in class ai.beestream.jitnn.Shape
The rank-0 shape: a scalar with one element.
scalar(float) - Static method in class ai.beestream.jitnn.Tensor
A rank-0 FP32 tensor holding value.
scale() - Method in class ai.beestream.jitnn.nn.optim.GradScaler
The current loss scale.
scale(Tensor) - Method in class ai.beestream.jitnn.nn.optim.GradScaler
loss * scale.
scaledDotProductAttention(Tensor, Tensor, Tensor, Tensor, boolean, float) - Static method in class ai.beestream.jitnn.F
Scaled dot-product attention, PyTorch semantics: softmax(q k^T * scale + mask) v, for q [..., T, E], k [..., S, E], v [..., S, Ev].
scaledDotProductAttentionBTHD(Tensor, Tensor, Tensor, Tensor, boolean, float) - Static method in class ai.beestream.jitnn.F
Scaled dot-product attention on q [B, T, H, E], k and v [B, S, H, E] (heads after tokens, as the projections produce them), returning [B, T, H, E]; otherwise as F.scaledDotProductAttention(Tensor, Tensor, Tensor, Tensor, boolean, float).
scatter(int, Tensor, Tensor) - Method in class ai.beestream.jitnn.Tensor
torch.scatter: this with src written at index (duplicates: the last wins); no gradient.
scatter(NDArray, NDArray, int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
scatterAdd(int, Tensor, Tensor) - Method in class ai.beestream.jitnn.Tensor
torch.scatter_add: this with src added at index along axis.
scatterND(Tensor, Tensor, Tensor, String) - Static method in class ai.beestream.jitnn.F
A copy of data with the update slices written ("none"), added ("add") or multiplied ("mul") at the INT32 index rows indices [..., k] (ONNX ScatterND); differentiable for none and add.
scratch(int) - Method in class ai.beestream.jitnn.engine.Emitter
A device work buffer owned by the compiled plan (partial results between two passes).
scratchF16(int) - Method in class ai.beestream.jitnn.engine.Emitter
An FP16 device work buffer owned by the compiled plan.
scratchI32(int) - Method in class ai.beestream.jitnn.engine.Emitter
An INT32 device work buffer owned by the compiled plan.
section() - Method in class ai.beestream.jitnn.JitnnConfig.Option
The section the documentation lists this option in.
seed(long) - Static method in class ai.beestream.jitnn.Tensor
A 64-bit seed as the INT32 [2] host tensor [lo, hi] that the seeded random operators take.
segmentReduce(Tensor, int[], F.BagMode) - Static method in class ai.beestream.jitnn.F
Sum, mean or max of the row segments of x [rows, D]: segment b covers rows offsets[b] .. offsets[b + 1] - 1, the last to the end; an empty segment gives zeros.
selu() - Method in class ai.beestream.jitnn.Tensor
SELU with PyTorch's constants (alpha 1.6732632, scale 1.0507010).
SELU - Class in ai.beestream.jitnn.nn
SELU with PyTorch's constants.
SELU() - Constructor for class ai.beestream.jitnn.nn.SELU
 
sentences() - Method in class ai.beestream.jitnn.nn.text.Corpus
The sentences, as word ids.
sequenceMask(NDArray, float) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
Sequential - Class in ai.beestream.jitnn.nn
Applies its modules in order.
Sequential(Module...) - Constructor for class ai.beestream.jitnn.nn.Sequential
The modules applied in order, each to the previous one's output.
session() - Method in class ai.beestream.jitnn.djl.JitnnSymbolBlock
The jitnn-onnx session behind this block.
session() - Method in class ai.beestream.jitnn.tf.TfModel
The jitnn-onnx session the model was translated to.
set(JitnnConfig.Option, T) - Method in class ai.beestream.jitnn.JitnnConfig.Builder
Sets option (null: back to its default).
set(NDIndex, NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
set(NDIndex, Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
set(NDIndex, Function) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
set(NDArray, Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
set(String, String) - Method in class ai.beestream.jitnn.JitnnConfig.Builder
Sets the option with this key from its text form, as a system property would.
setAutotune(String) - Static method in class ai.beestream.jitnn.JitnnRuntime
Sets run-time autotuning: "on" measures new shapes, "use" only applies saved decisions, "off".
setAutotuneDirectory(Path) - Static method in class ai.beestream.jitnn.JitnnRuntime
Sets where autotuning decisions are saved; decisions loaded so far are dropped.
setBackend(Backend) - Static method in class ai.beestream.jitnn.JitnnRuntime
Switches this thread's backend, first running anything recorded on the old one.
setDeterministic(boolean) - Static method in class ai.beestream.jitnn.JitnnRuntime
Run-to-run reproducible results, as PyTorch's use_deterministic_algorithms: jitNN's own kernels are deterministic already (no atomics; split reductions sum in a fixed order); this also stops autotuning from measuring and asks cuDNN for deterministic algorithms only.
setGrad(Tensor) - Method in class ai.beestream.jitnn.Tensor
Replaces the gradient of this leaf tensor (same shape), for example by its average over data-parallel ranks before the optimiser step.
setLearningRate(float) - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Changes the learning rate; takes effect at the next step without recompiling.
setRandomSeed(int) - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
setRequiresGradient(boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
setScalar(NDIndex, Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
SGD - Class in ai.beestream.jitnn.nn.optim
Stochastic gradient descent with optional (Nesterov) momentum and L2 weight decay, as in PyTorch.
SGD(List, float) - Constructor for class ai.beestream.jitnn.nn.optim.SGD
Plain SGD: no momentum, no weight decay.
SGD(List, float, float, float, boolean) - Constructor for class ai.beestream.jitnn.nn.optim.SGD
SGD with the given momentum (0 for none), L2 weight decay, and Nesterov momentum when nesterov is set.
shape() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns the value of the shape record component.
shape() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns the value of the shape record component.
shape() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
Returns the value of the shape record component.
shape() - Method in class ai.beestream.jitnn.Tensor
The shape of this tensor.
shape(String) - Method in class ai.beestream.jitnn.io.StateDict
The shape of tensor name.
Shape - Class in ai.beestream.jitnn
An immutable, row-major tensor shape.
sharedScratch(String, int) - Method in class ai.beestream.jitnn.engine.Emitter
A work buffer of exactly numel elements shared by every task of the plan that asks for the same role and size: tasks run in order, so a buffer one operator fills and reads is free for the next (the layout copies around library calls, which would otherwise take a buffer each).
sharedScratchF16(String, int) - Method in class ai.beestream.jitnn.engine.Emitter
shiftLeft(Tensor) - Method in class ai.beestream.jitnn.Tensor
this << o for INT32 tensors, with broadcasting.
shiftRight(Tensor) - Method in class ai.beestream.jitnn.Tensor
this >> o (arithmetic: the sign is kept) for INT32 tensors, with broadcasting.
shuffle(long) - Method in class ai.beestream.jitnn.data.DataLoader.Builder
Reshuffles every epoch, from seed and the epoch number.
sigmoid() - Method in class ai.beestream.jitnn.Tensor
The logistic sigmoid 1 / (1 + exp(-x)), element-wise.
Sigmoid - Class in ai.beestream.jitnn.nn
The logistic sigmoid 1 / (1 + exp(-x)).
Sigmoid() - Constructor for class ai.beestream.jitnn.nn.Sigmoid
 
sign() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sign() - Method in class ai.beestream.jitnn.Tensor
Element-wise sign: -1, 0 or 1.
signi() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
silu() - Method in class ai.beestream.jitnn.Tensor
SiLU (swish): x * sigmoid(x).
sin() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sin() - Method in class ai.beestream.jitnn.Tensor
Element-wise sine.
sinh() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sinh() - Method in class ai.beestream.jitnn.Tensor
Element-wise hyperbolic sine.
size() - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Returns the value of the size record component.
size() - Method in class ai.beestream.jitnn.nn.text.Vocabulary
The number of words.
size(int) - Method in class ai.beestream.jitnn.Buckets
The smallest size that holds n; throws if n is larger than every size.
sizes() - Method in class ai.beestream.jitnn.Buckets
The sizes, ascending.
skip() - Method in class ai.beestream.jitnn.onnx.Proto
Skips the current field's value.
SKIP_GRAM - Enum constant in enum class ai.beestream.jitnn.nn.models.Word2Vec.Mode
The centre word predicts each context word.
skipGramPairs(int, boolean) - Method in class ai.beestream.jitnn.nn.text.Corpus
Skip-gram (centre, context) pairs with word2vec's dynamic window (each centre draws a window 1..window), shuffled: [centres, contexts].
slice(int, int, int) - Method in class ai.beestream.jitnn.Tensor
Elements start .. end - 1 along axis; negative indices count from the end.
slogdet(Tensor) - Static method in class ai.beestream.jitnn.Linalg
torch.linalg.slogdet: [sign, log |det|], each [...]; only the logarithm carries a gradient.
smoothL1Loss(Tensor, Tensor, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Smooth L1 loss, PyTorch's smooth_l1_loss: Huber with delta = beta, divided by beta.
softmax() - Method in class ai.beestream.jitnn.Tensor
Softmax along the last axis.
softmax(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
softplus() - Method in class ai.beestream.jitnn.Tensor
Softplus, log(1 + exp(x)) (beta 1), computed without overflow.
Softplus - Class in ai.beestream.jitnn.nn
Softplus: log(1 + exp(x)).
Softplus() - Constructor for class ai.beestream.jitnn.nn.Softplus
 
softsign() - Method in class ai.beestream.jitnn.Tensor
Softsign: x / (1 + |x|).
Softsign - Class in ai.beestream.jitnn.nn
Softsign: x / (1 + |x|).
Softsign() - Constructor for class ai.beestream.jitnn.nn.Softsign
 
solve(Tensor, Tensor) - Static method in class ai.beestream.jitnn.Linalg
Solves A X = B for square A [..., n, n] and B [..., n, k] through LU with partial pivoting.
solveTriangular(Tensor, Tensor, boolean, boolean) - Static method in class ai.beestream.jitnn.Linalg
Solves T X = B for triangular T, torch.linalg.solve_triangular (left side): only T's upper or lower triangle is read, and with unitDiagonal not its diagonal either.
sort() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sort(int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sort(int, boolean) - Method in class ai.beestream.jitnn.Tensor
Stable sort along axis: {values, INT32 original indices}; values carry the gradient.
Source(DType, long[], MemorySegment) - Constructor for record class ai.beestream.jitnn.io.SafeTensors.Source
Creates an instance of a Source record class.
spaceToDepth(Tensor, int) - Static method in class ai.beestream.jitnn.F
Moves blocks of space into channels: x [N, C, H, W] to [N, C b^2, H / b, W / b] (ONNX SpaceToDepth; inverse of DCR depthToSpace).
SparseMatrix - Class in ai.beestream.jitnn
The structure of a sparse matrix (which entries are stored), built once on the host from coordinate lists; the values are a separate tensor, in the order the coordinates were given, so they can be learned.
SparseMatrix.Valued - Record Class in ai.beestream.jitnn
A structure with fixed values.
sparsemax(Tensor) - Static method in class ai.beestream.jitnn.F
Sparsemax (Martins and Astudillo, 2016) along the last axis: the Euclidean projection onto the probability simplex, max(x - tau, 0) with tau set by the sorted support.
split(int, int...) - Method in class ai.beestream.jitnn.Tensor
Splits axis into consecutive pieces of the given sizes.
split(long[], int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
split(long, int) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
SPLIT_K - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.splitK.
SPLIT_K_GROUPS - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.splitK.groups.
sqrt() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sqrt() - Method in class ai.beestream.jitnn.Tensor
The square root, element-wise.
square() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
square() - Method in class ai.beestream.jitnn.Tensor
x^2, element-wise.
squaredHingeLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
Squared hinge loss: max(0, 1 - target * input)^2.
squeeze() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
squeeze(int) - Method in class ai.beestream.jitnn.Tensor
A view without the size-1 axis axis.
squeeze(int[]) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
stack(int, Tensor...) - Static method in class ai.beestream.jitnn.Tensor
Joins same-shape tensors along a new axis.
startAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int) - Method in interface ai.beestream.jitnn.dist.Collective
Starts the reduction over all ranks of tensors, packed into the persistent buffer (exactly their total size), so that it runs while the operators recorded after it do: NCCL runs it on a stream of its own.
startAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int) - Method in class ai.beestream.jitnn.dist.NcclCollective
Packs tensors into persistent buffer and starts its all-reduce on NCCL's side stream (see Collective.startAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int)).
startTrace() - Static method in class ai.beestream.jitnn.JitnnRuntime
Starts recording a timeline of plan executions for JitnnRuntime.stopTrace(Path).
state() - Method in class ai.beestream.jitnn.nn.optim.AdaDelta
 
state() - Method in class ai.beestream.jitnn.nn.optim.AdaGrad
 
state() - Method in class ai.beestream.jitnn.nn.optim.Adam
 
state() - Method in class ai.beestream.jitnn.nn.optim.AdaMax
 
state() - Method in class ai.beestream.jitnn.nn.optim.Lamb
 
state() - Method in class ai.beestream.jitnn.nn.optim.Lars
 
state() - Method in class ai.beestream.jitnn.nn.optim.NAdam
 
state() - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Per-parameter state by slot name, each array indexed like the parameters (null where none).
state() - Method in class ai.beestream.jitnn.nn.optim.RMSProp
 
state() - Method in class ai.beestream.jitnn.nn.optim.SGD
 
stateDict() - Method in class ai.beestream.jitnn.nn.Module
The parameters (parameters.<i>) and buffers (buffers.<i>) in registration order, referring to the live tensors: save it with StateDict.save(Path).
stateDict() - Method in class ai.beestream.jitnn.nn.optim.GradScaler
The scale and the count of clean steps since it last changed, for checkpoints.
stateDict() - Method in class ai.beestream.jitnn.nn.optim.LRScheduler
The epoch (or step) count, for checkpoints; the rate itself is restored with the optimiser.
stateDict() - Method in class ai.beestream.jitnn.nn.optim.Optimizer
The optimiser's state for a checkpoint: its type, learning rate, step count and per-parameter state (state.<slot>.<parameter index>), which refers to the live tensors.
StateDict - Class in ai.beestream.jitnn.io
Named tensors and string values: the state of a model, an optimiser or a whole training run, as PyTorch's state_dict.
StateDict() - Constructor for class ai.beestream.jitnn.io.StateDict
An empty dictionary, to fill with put and save.
step() - Method in class ai.beestream.jitnn.nn.optim.AdaDelta
 
step() - Method in class ai.beestream.jitnn.nn.optim.AdaGrad
 
step() - Method in class ai.beestream.jitnn.nn.optim.Adam
 
step() - Method in class ai.beestream.jitnn.nn.optim.AdaMax
 
step() - Method in class ai.beestream.jitnn.nn.optim.Lamb
 
step() - Method in class ai.beestream.jitnn.nn.optim.Lars
 
step() - Method in class ai.beestream.jitnn.nn.optim.LRScheduler
Advances the count and sets the optimiser's rate for it.
step() - Method in class ai.beestream.jitnn.nn.optim.NAdam
 
step() - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Records one update of every parameter that has a gradient.
step() - Method in class ai.beestream.jitnn.nn.optim.RMSProp
 
step() - Method in class ai.beestream.jitnn.nn.optim.SGD
 
step(Optimizer) - Method in class ai.beestream.jitnn.nn.optim.GradScaler
Unscales (unless done already) and records the optimiser step, skipped on the device on an overflow.
step(Optimizer, int, float) - Static method in class ai.beestream.jitnn.nn.optim.LRScheduler
lr = base * gamma^(epoch / stepSize).
stft(Tensor, int, int, Tensor, boolean, boolean) - Static method in class ai.beestream.jitnn.Fft
torch.stft(x, nFft, hop, window=window, center, pad_mode="reflect", normalized, onesided=True, return_complex=True) as real pairs: x [L] or [B, L] gives [(B,) nFft / 2 + 1, frames, 2].
stopGradient() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
stopTrace(Path) - Static method in class ai.beestream.jitnn.JitnnRuntime
Stops recording and writes the timeline as Chrome trace events (JSON), for Perfetto or chrome://tracing: plan executions, kernels, and the spans opened with JitnnRuntime.traceRange(String).
storage() - Method in class ai.beestream.jitnn.Tensor
The memory behind this tensor, shared with its views.
strides() - Method in class ai.beestream.jitnn.Shape
Contiguous row-major strides, in elements.
string() - Method in class ai.beestream.jitnn.onnx.Proto
The current length-delimited field as a UTF-8 string.
string(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
A string value; throws when the key is absent or holds another type.
strings(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
An array of strings (a tokenizer's vocabulary, say); throws when the key is absent or holds another type.
sub(Tensor) - Method in class ai.beestream.jitnn.Tensor
this - o, element-wise with broadcasting.
sub(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sub(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sub(String) - Method in class ai.beestream.jitnn.io.StateDict
The entries under prefix., with the prefix removed.
subi(NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
subi(Number) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sum() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sum() - Method in class ai.beestream.jitnn.Tensor
The sum of all elements, as a scalar.
sum(int[], boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
sum(int, boolean) - Method in class ai.beestream.jitnn.Tensor
The sum along axis, which is kept with size 1 when keepDim is set.
SUM - Enum constant in enum class ai.beestream.jitnn.dist.Collective.ReduceOp
The sum over ranks.
SUM - Enum constant in enum class ai.beestream.jitnn.F.BagMode
The sum of the rows.
SUM - Enum constant in enum class ai.beestream.jitnn.F.Reduction
The sum of the per-element losses.
sumTo(Shape) - Method in class ai.beestream.jitnn.Tensor
Sums broadcast dimensions away so the result has target shape (gradient of broadcasting).
svd(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Reduced SVD, torch.linalg.svd(full_matrices=False): U [..., m, k], S [..., k] descending and V [..., n, k] (V, not V^T), k = min(m, n).
svdvals(Tensor) - Static method in class ai.beestream.jitnn.Linalg
Singular values (descending), torch.linalg.svdvals.
swiglu(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
SwiGLU gating, silu(gate) * up, as in the Llama feed-forward block.
sync(Tensor...) - Static method in class ai.beestream.jitnn.JitnnRuntime
Runs everything recorded on this thread, fetching wanted to the host.
SyncBatchNorm2d - Class in ai.beestream.jitnn.nn
Batch normalisation whose training statistics are those of the batch over all data-parallel ranks, as PyTorch's SyncBatchNorm: each rank sums its x and x^2 per channel, one all-reduce gives the global mean and variance, and the backward pass all-reduces their gradients the same way.
SyncBatchNorm2d(int, Collective, int) - Constructor for class ai.beestream.jitnn.nn.SyncBatchNorm2d
Synchronised batch normalisation for rank rank, with PyTorch's defaults (momentum 0.1, eps 1e-5).
SyncBatchNorm2d(int, Collective, int, float, float) - Constructor for class ai.beestream.jitnn.nn.SyncBatchNorm2d
Synchronised batch normalisation for rank rank of an NcclCollective; with a null collective (or one rank) it is plain batch normalisation.

T

t - Variable in class ai.beestream.jitnn.nn.optim.Optimizer
Steps taken (the bias corrections of the adaptive methods use it).
t() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the t record component.
t() - Method in class ai.beestream.jitnn.Tensor
Transpose of a matrix.
take(NDManager, NDArray) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
takeStats() - Method in class ai.beestream.jitnn.nn.optim.Optimizer
The statistics the next update uses (computed now when a pre-scale is pending), or null; clears them.
tan() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
tan() - Method in class ai.beestream.jitnn.Tensor
Element-wise tangent.
tanh() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
tanh() - Method in class ai.beestream.jitnn.Tensor
The hyperbolic tangent, element-wise.
Tanh - Class in ai.beestream.jitnn.nn
The hyperbolic tangent.
Tanh() - Constructor for class ai.beestream.jitnn.nn.Tanh
 
task(WorkerGrid, Emitter.TaskAdder) - Method in class ai.beestream.jitnn.engine.Emitter
Adds a kernel task launched over grid; adder adds it to the graph, with arrays from this emitter's accessors.
tensor() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
The jitNN tensor behind this array (FP32).
tensor(String) - Method in class ai.beestream.jitnn.gguf.GgufFile
Tensor name as a persistent jitNN tensor, memory-mapped where possible (see the class comment).
tensor(String) - Method in class ai.beestream.jitnn.io.StateDict
Tensor name as a new persistent tensor (a weight that stays on the device).
tensor(String, boolean) - Method in class ai.beestream.jitnn.gguf.GgufFile
Tensor name; copy reads it into memory instead of mapping it.
Tensor - Class in ai.beestream.jitnn
An n-dimensional array.
TENSOR_CORES - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.tensorCores.
TensorData(String, long[], int, ByteBuffer) - Constructor for record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Creates an instance of a TensorData record class.
tensorInfos() - Method in class ai.beestream.jitnn.gguf.GgufFile
The tensors listed in the file, by name, in file order.
tensorNames() - Method in class ai.beestream.jitnn.io.StateDict
The tensor names, in insertion (or file) order.
tensors() - Method in class ai.beestream.jitnn.gguf.GgufFile
Every tensor in the file (all must be of a type jitNN loads), by name, in file order.
TextCorpus - Class in ai.beestream.jitnn.data
A byte-level text corpus for language-model gates: every byte is a token (vocabulary 256).
TfModel - Class in ai.beestream.jitnn.tf
A TensorFlow model on jitNN: a SavedModel directory (TensorFlow 2 / Keras export, with its variables checkpoint) or a frozen GraphDef file.
then(Transforms.Transform) - Method in interface ai.beestream.jitnn.data.Transforms.Transform
This transform, then next.
theta() - Method in class ai.beestream.jitnn.ml.NaiveBayes
Gaussian: the per-class feature means [classes, features].
thresholdedRelu(float) - Method in class ai.beestream.jitnn.Tensor
ThresholdedReLU (Keras): x where x > theta, else 0.
ThresholdedReLU - Class in ai.beestream.jitnn.nn
Keras's ThresholdedReLU: x where x > theta, else 0.
ThresholdedReLU(float) - Constructor for class ai.beestream.jitnn.nn.ThresholdedReLU
ThresholdedReLU with threshold theta.
tile(int...) - Method in class ai.beestream.jitnn.Tensor
This tensor repeated reps[i] times along each axis (torch.Tensor.repeat, numpy.tile, ONNX Tile); reps longer than the rank adds leading axes.
tile(int, long) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
tile(long) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
tile(long[]) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
tile(Shape) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
timeSplit() - Static method in class ai.beestream.jitnn.JitnnRuntime
Cumulative nanoseconds spent in jitNN's executor outside TornadoVM ([0]) and inside TornadoVM's plan execution ([1]).
title() - Method in enum class ai.beestream.jitnn.JitnnConfig.Section
The heading this section has in the documentation.
to(DType) - Method in class ai.beestream.jitnn.Tensor
This tensor converted to dtype: between FP32, FP16 and BF16 (round to nearest even; differentiable), or between FP32 and INT32 (truncating toward zero; no gradient).
toByteBuffer(boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
toDegrees() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
toDevice(Device, boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
toFloat() - Method in class ai.beestream.jitnn.Tensor
This tensor converted to FP32 (to(DType.F32)).
toFloatArray() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
toFloatArray() - Method in class ai.beestream.jitnn.Tensor
The elements in row-major order, converted to float from any element type (INT32, FP16, BF16, FP8, GGUF blocks).
toFloatArray(float[]) - Method in class ai.beestream.jitnn.Tensor
Copies an F32, F16 or BF16 tensor into dst (length Tensor.numel()) without allocating: for BERT's [32, 128, 768] output, a fresh float[] costs more than the copy itself.
toIntArray() - Method in class ai.beestream.jitnn.Tensor
The elements of an INT32 tensor in row-major order.
token(int) - Method in class ai.beestream.jitnn.data.TextCorpus
Token i.
token(int) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
The token with id id, or null.
tokenize(String) - Method in class ai.beestream.jitnn.nn.text.Tokenizer
The ids of text without special tokens.
Tokenizer - Class in ai.beestream.jitnn.nn.text
A Hugging Face tokenizer read from its tokenizer.json, for the two families jitNN's models use: WordPiece with the BERT normaliser and pre-tokeniser (BERT and its relatives), and byte-level BPE (GPT-2, RoBERTa).
Tokenizer.Batch - Record Class in ai.beestream.jitnn.nn.text
Encodings padded to the longest one: [batch, length] ids and type ids, and an attention mask of 1 for tokens and 0 for padding.
Tokenizer.Encoding - Record Class in ai.beestream.jitnn.nn.text
The ids, type ids and tokens of one text (or pair), special tokens included.
tokens() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Returns the value of the tokens record component.
topk(int) - Method in class ai.beestream.jitnn.Tensor
The k largest values along the last axis and their INT32 indices (descending; ties: lower index first).
topK(int, int, boolean, boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
toProperties() - Method in class ai.beestream.jitnn.JitnnConfig
The options that were set, as key=value properties.
toRadians() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
TORNADO - Enum constant in enum class ai.beestream.jitnn.Backend
Operators are recorded lazily; a sync point compiles everything recorded since the last one into one TornadoVM task graph, caches the compiled plan by the trace's structure, and replays it when the same structure comes back (a training step).
toString() - Method in class ai.beestream.jitnn.Buckets
 
toString() - Method in class ai.beestream.jitnn.CustomOp
 
toString() - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Returns a string representation of this record class.
toString() - Method in class ai.beestream.jitnn.djl.JitnnEngine
 
toString() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
toString() - Method in class ai.beestream.jitnn.engine.Node
 
toString() - Method in class ai.beestream.jitnn.gguf.GgufFile
 
toString() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.io.SafeTensors.Header
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
Returns a string representation of this record class.
toString() - Method in class ai.beestream.jitnn.JitnnConfig.Option
 
toString() - Method in class ai.beestream.jitnn.JitnnConfig
 
toString() - Method in record class ai.beestream.jitnn.nn.GRU.Output
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.nn.models.Vae.Output
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.RopeScaling.Linear
Returns a string representation of this record class.
toString() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
Returns a string representation of this record class.
toString() - Method in class ai.beestream.jitnn.Shape
 
toString() - Method in record class ai.beestream.jitnn.SparseMatrix.Valued
Returns a string representation of this record class.
toString() - Method in class ai.beestream.jitnn.Tensor
 
toType(DataType, boolean) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
traceRange(String) - Static method in class ai.beestream.jitnn.JitnnRuntime
A named span on the timeline (one training step, say) while recording; nothing otherwise.
trafficStats() - Static method in class ai.beestream.jitnn.JitnnRuntime
Totals over every plan execution so far: [executions, kernel launches, library calls, bytes], the bytes being what the operators read and write at least (a lower bound on device-memory traffic that fusion lowers).
train(boolean) - Method in class ai.beestream.jitnn.nn.Module
Switches this module and its children to training mode (on) or evaluation mode; returns this module.
trainBatch(int, int, int, int[], int[]) - Method in class ai.beestream.jitnn.data.TextCorpus
Fills inputs and next-token targets [rows, seq] for training batch b.
trainStart(int, int, int, int) - Method in class ai.beestream.jitnn.data.TextCorpus
Start of training row r of batch b: consecutive windows, wrapping around the training part.
transferStats() - Static method in class ai.beestream.jitnn.JitnnRuntime
Host-device traffic of this JVM's plans so far: [fetches to the host, elements fetched, re-uploads of persistent storages written by another plan, elements re-uploaded, input copies, elements copied].
transform(Transforms.Transform) - Method in class ai.beestream.jitnn.data.DataLoader.Builder
Applied to every sample as its batch is assembled (augmentation, normalisation).
transform(Tensor) - Method in class ai.beestream.jitnn.ml.Pca
Projection [n, k] of x onto the components.
Transforms - Class in ai.beestream.jitnn.data
Per-sample transforms for DataLoader, applied in place to one CHW sample on the host: the usual image augmentations and normalisation.
Transforms.Transform - Interface in ai.beestream.jitnn.data
Transforms one sample of shape at x[at ..] in place.
transpose() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
transpose(int...) - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
transpose(int, int) - Method in class ai.beestream.jitnn.Tensor
Axes a and b swapped (negative axes count from the end); a copy, as Tensor.permute(int...).
TRANSPOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.transpose.
tree(RandomForest.Task, int, int) - Static method in class ai.beestream.jitnn.ml.RandomForest
One decision tree.
tripletMarginLoss(Tensor, Tensor, Tensor, float, F.Reduction) - Static method in class ai.beestream.jitnn.F
Triplet margin loss: max(d(anchor, positive) - d(anchor, negative) + margin, 0).
trunc() - Method in class ai.beestream.jitnn.djl.JitnnNDArray
 
trunc() - Method in class ai.beestream.jitnn.Tensor
Element-wise rounding towards zero.
truncatedNormal(float, float, Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
Tsne - Class in ai.beestream.jitnn.ml
Exact t-SNE (van der Maaten and Hinton, 2008) on the device, after scikit-learn's TSNE with method="exact": Gaussian affinities calibrated per point to the perplexity by bisection on the precision, symmetrised; a Student-t kernel in the embedding; gradient descent with momentum (0.5, then 0.8), early exaggeration (12 for 250 iterations), per-parameter gains (+0.2 / x0.8, at least 0.01) and the learning rate max(n / 12 / 4, 50).
Tsne() - Constructor for class ai.beestream.jitnn.ml.Tsne
t-SNE with scikit-learn's defaults: 2 dimensions, perplexity 30, 1000 iterations.
Tsne(int, float, int) - Constructor for class ai.beestream.jitnn.ml.Tsne
t-SNE into components dimensions with the given perplexity and number of iterations.
TUNING - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.tuning.
type() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns the value of the type record component.
type() - Method in class ai.beestream.jitnn.JitnnConfig.Option
The value type: Boolean, Integer, Long, Double or String.
type() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
Returns the value of the type record component.
typeId() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
Returns the value of the typeId record component.
typeIdArray() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
Returns the value of the typeIdArray record component.
typeIds() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
The token type (segment) ids as an INT32 tensor [batch, length].
typeIds() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
Returns the value of the typeIds record component.
typeName() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
Returns the value of the typeName record component.

U

UINT16 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 16-bit unsigned integer.
UINT8 - Static variable in class ai.beestream.jitnn.onnx.OnnxModel
ONNX data type: 8-bit unsigned integer.
unary(String, TornadoFunctions.Task4, DoubleUnaryOperator) - Static method in class ai.beestream.jitnn.CustomOp
An element-wise operator on one FP32 tensor.
UNet - Class in ai.beestream.jitnn.nn.models
A small U-Net (Ronneberger et al., 2015) for 28x28 inputs with two down-sampling levels: double 3x3 convolutions with ReLU, max pooling, a bottleneck, then one level up through a stride-2 transposed convolution and one through bilinear upsampling and a 1x1 convolution, each concatenated with the encoder's features of that size; a 1x1 convolution gives one logit per pixel.
UNet(int, int, long) - Constructor for class ai.beestream.jitnn.nn.models.UNet
A U-Net over images of in channels, width channels at the first level (doubling at each level down), initialised from seed.
uniform(long, float, float, int...) - Static method in class ai.beestream.jitnn.Tensor
Uniform in [lo, hi), from a counter-based generator: the same seed gives the same values everywhere.
uniform(Tensor, float, float, int...) - Static method in class ai.beestream.jitnn.Tensor
As Tensor.uniform(long, float, float, int...) with the seed in an INT32 [2] tensor (see Tensor.seed(long)): a plan replayed with a new seed draws new values without recompiling, so use this for noise drawn every step.
unique(Tensor, boolean, boolean) - Static method in class ai.beestream.jitnn.F
torch.unique (sorted): the distinct values of x, then (when asked for, else null) the INT32 index of each element's value among them, shaped like x, and the INT32 count of each value.
uniqueId() - Static method in class ai.beestream.jitnn.dist.NcclCollective
A new NCCL unique id (base64) for NcclCollective(String, int, int, String...); this process then hosts NCCL's rendezvous.
unread(Tensor) - Method in class ai.beestream.jitnn.engine.Emitter
True when nothing reads t after this plan: an internal output (no user code holds it) that no later operator of the trace consumes and that was not requested.
unscale(Optimizer) - Method in class ai.beestream.jitnn.nn.optim.GradScaler
Makes opt divide its gradients by the scale at the next step (and clip after unscaling).
unsqueeze(int) - Method in class ai.beestream.jitnn.Tensor
A view with a new size-1 axis at axis (negative counts from the end of the result).
unused() - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
Weights of the checkpoint that no load call read, e.g. a head the model does not have.
update() - Method in class ai.beestream.jitnn.nn.optim.GradScaler
Adjusts the scale from this step's overflow flag (reading it waits for the step) and returns whether the step's update was skipped.
Upsample - Class in ai.beestream.jitnn.nn
Nearest or bilinear upsampling of NCHW by a scale factor, PyTorch's nn.Upsample.
Upsample(float, boolean, boolean) - Constructor for class ai.beestream.jitnn.nn.Upsample
Upsampling by scale, nearest or bilinear; alignCorners as in PyTorch (bilinear only).
upsampleBilinear(Tensor, float, boolean) - Static method in class ai.beestream.jitnn.F
Bilinear upsampling of x [N, C, H, W] by scale.
upsampleNearest(Tensor, float) - Static method in class ai.beestream.jitnn.F
Nearest-neighbour upsampling of x [N, C, H, W] by scale (PyTorch's mode='nearest').

V

Vae - Class in ai.beestream.jitnn.nn.models
A fully connected variational autoencoder (Kingma and Welling, 2014), the layout of PyTorch's MNIST example: 784-400 encoder to a Gaussian latent (mean, log-variance), 400-784 decoder with a sigmoid output.
Vae(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.models.Vae
An autoencoder of in features through hidden to a latent of latent values, initialised from seed.
Vae.Output - Record Class in ai.beestream.jitnn.nn.models
Reconstruction probabilities [B, in] and the latent mean and log-variance [B, latent].
value(String) - Method in class ai.beestream.jitnn.io.StateDict
The value name; throws when it is missing.
Valued(SparseMatrix, Tensor) - Constructor for record class ai.beestream.jitnn.SparseMatrix.Valued
Creates an instance of a Valued record class.
ValueInfo(String, int, long[]) - Constructor for record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
A value without names for its symbolic dimensions.
ValueInfo(String, int, long[], String[]) - Constructor for record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
Creates an instance of a ValueInfo record class.
valueNames() - Method in class ai.beestream.jitnn.io.StateDict
The value names.
valueOf(String) - Static method in enum class ai.beestream.jitnn.Backend
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.dist.Collective.ReduceOp
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.DType
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.F.BagMode
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.F.PadMode
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.F.PoolMode
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.F.Reduction
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.Fft.Norm
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.gguf.GgmlType
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.JitnnConfig.Section
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.ml.NaiveBayes.Kind
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.ml.RandomForest.Task
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.nn.models.Doc2Vec.Mode
Returns the enum constant of this class with the specified name.
valueOf(String) - Static method in enum class ai.beestream.jitnn.nn.models.Word2Vec.Mode
Returns the enum constant of this class with the specified name.
values() - Static method in enum class ai.beestream.jitnn.Backend
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.dist.Collective.ReduceOp
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.DType
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.F.BagMode
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.F.PadMode
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.F.PoolMode
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.F.Reduction
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.Fft.Norm
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.gguf.GgmlType
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.JitnnConfig.Section
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.ml.NaiveBayes.Kind
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.ml.RandomForest.Task
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.nn.models.Doc2Vec.Mode
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Static method in enum class ai.beestream.jitnn.nn.models.Word2Vec.Mode
Returns an array containing the constants of this enum class, in the order they are declared.
values() - Method in record class ai.beestream.jitnn.SparseMatrix.Valued
Returns the value of the values record component.
var() - Method in class ai.beestream.jitnn.ml.NaiveBayes
Gaussian: the per-class feature variances [classes, features], smoothed.
varint() - Method in class ai.beestream.jitnn.onnx.Proto
Reads a varint.
vectors() - Method in class ai.beestream.jitnn.nn.models.GloVe
All word vectors [vocab, dim] on the host.
vectors() - Method in class ai.beestream.jitnn.nn.models.Word2Vec
All word vectors [vocab, dim] on the host.
VERBOSE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.verbose.
version() - Method in class ai.beestream.jitnn.gguf.GgufFile
The GGUF format version (2 or 3).
VOCAB - Static variable in class ai.beestream.jitnn.data.TextCorpus
The vocabulary size: one token per byte value.
vocabSize() - Method in class ai.beestream.jitnn.nn.text.Tokenizer
The number of tokens in the vocabulary.
Vocabulary - Class in ai.beestream.jitnn.nn.text
Word ids by descending frequency, the words seen at least minCount times.

W

warmupCosine(Optimizer, int, int, float) - Static method in class ai.beestream.jitnn.nn.optim.LRScheduler
Linear warm-up over warmup steps, then a half-cosine decay to minRatio times the base rate at total steps (Hugging Face's get_cosine_schedule_with_warmup when minRatio is 0).
warmupLinear(Optimizer, int, int) - Static method in class ai.beestream.jitnn.nn.optim.LRScheduler
Linear warm-up from 0 to the base rate over warmup steps, then linear decay to 0 at total steps (Hugging Face's get_linear_schedule_with_warmup).
weight() - Method in class ai.beestream.jitnn.nn.Conv1d
The weight [out, in, kernel].
weight() - Method in class ai.beestream.jitnn.nn.Conv2d
The weight [out, in/groups, kernel, kernel].
weight() - Method in class ai.beestream.jitnn.nn.Conv3d
The weight [out, in, kernel, kernel, kernel].
weight() - Method in class ai.beestream.jitnn.nn.ConvTranspose2d
The weight [in, out/groups, kernel, kernel].
weight() - Method in class ai.beestream.jitnn.nn.Embedding
The table [count, dim].
weight() - Method in class ai.beestream.jitnn.nn.EmbeddingBag
The table [count, dim].
weight() - Method in class ai.beestream.jitnn.nn.GraphConv
The weight [out, in].
weight() - Method in class ai.beestream.jitnn.nn.LayerNorm
The scale [features].
weight() - Method in class ai.beestream.jitnn.nn.Linear
The weight [out, in].
weight() - Method in class ai.beestream.jitnn.nn.PReLU
The slopes [parameters].
weight() - Method in class ai.beestream.jitnn.nn.RMSNorm
The scale [features].
weightNorm(Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
Weight normalisation (Salimans and Kingma): w = g v / ||v|| with the norm over every axis but 0, g of shape [out] (PyTorch's weight_norm with dim 0).
where(Tensor, Tensor, Tensor) - Static method in class ai.beestream.jitnn.F
cond !
WINOGRAD - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.winograd.
wire() - Method in class ai.beestream.jitnn.onnx.Proto
The wire type of the current field (0 varint, 1 fixed64, 2 length-delimited, 5 fixed32).
with(Backend, Supplier) - Static method in class ai.beestream.jitnn.JitnnRuntime
Runs body on backend, syncs, and restores the previous backend.
with(JitnnConfig.Option, T) - Method in class ai.beestream.jitnn.JitnnConfig
A copy with option set to value (null: back to its default).
WITH_NATIVE - Static variable in class ai.beestream.jitnn.JitnnConfig
jitnn.withNative.
withBuckets(int...) - Method in class ai.beestream.jitnn.onnx.OnnxSession
Pads the model's symbolic dimensions to the smallest of sizes that holds them (a size beyond the largest is run as it is), so a server compiles one plan per bucket instead of one per request shape.
withGradient(CustomOp.BinaryGradient) - Method in class ai.beestream.jitnn.CustomOp
This element-wise binary operator with a gradient.
withGradient(CustomOp.UnaryGradient) - Method in class ai.beestream.jitnn.CustomOp
This element-wise unary operator with a gradient.
word(int) - Method in class ai.beestream.jitnn.nn.text.Vocabulary
The word with id id.
Word2Vec - Class in ai.beestream.jitnn.nn.models
word2vec (Mikolov et al., 2013) with negative sampling: an input embedding (the word vectors) and an output embedding, initialised like the original C code (input uniform in +-0.5 / dim, output zero).
Word2Vec(int, int, Word2Vec.Mode, long) - Constructor for class ai.beestream.jitnn.nn.models.Word2Vec
word2vec over vocab words with vectors of dim, trained as mode says, initialised from seed.
Word2Vec.Mode - Enum Class in ai.beestream.jitnn.nn.models
Which word predicts which.
workers(int) - Method in class ai.beestream.jitnn.data.DataLoader.Builder
Threads assembling batches ahead (0: on the caller's thread, when asked for).
worldSize() - Method in interface ai.beestream.jitnn.dist.Collective
The number of ranks, over every process.
worldSize() - Method in class ai.beestream.jitnn.dist.HostCollective
 
worldSize() - Method in class ai.beestream.jitnn.dist.NcclCollective
 
worldSize() - Method in class ai.beestream.jitnn.nn.DataParallel
The number of ranks.
wrap(float[], int...) - Static method in class ai.beestream.jitnn.Tensor
A tensor over data itself, without a copy, as PyTorch's from_numpy: the array is read when the operators that use it run (at the next sync), so it must not change until then, and later reads of the tensor see later changes.
wrap(int[], int...) - Static method in class ai.beestream.jitnn.Tensor
wrap(Tensor, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
Wraps a jitNN tensor as an array of this manager.
wrap(String, RuntimeException) - Static method in exception class ai.beestream.jitnn.JitnnException
e as a JitnnException when it is a device or runtime failure (one from TornadoVM), else e itself.
write(Path, Map, Map) - Static method in class ai.beestream.jitnn.io.SafeTensors
Writes tensors, in iteration order, and metadata to file.

X

x - Variable in class ai.beestream.jitnn.data.Blobs
The samples [count, dims], row-major.
x() - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Returns the value of the x record component.

Y

y - Variable in class ai.beestream.jitnn.data.Blobs
The cluster of each sample.
y() - Method in record class ai.beestream.jitnn.data.DataLoader.Batch
Returns the value of the y record component.
y() - Method in record class ai.beestream.jitnn.nn.GRU.Output
Returns the value of the y record component.
y() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
Returns the value of the y record component.

Z

zeroGrad() - Method in class ai.beestream.jitnn.nn.Module
Clears the gradients of every parameter.
zeroGrad() - Method in class ai.beestream.jitnn.nn.optim.Optimizer
Clears the gradients of every parameter.
zeroGrad() - Method in class ai.beestream.jitnn.Tensor
Clears the accumulated gradient of this leaf tensor.
zeros(int...) - Static method in class ai.beestream.jitnn.Tensor
An FP32 tensor of zeros.
zeros(Shape, DataType) - Method in class ai.beestream.jitnn.djl.JitnnNDManager
 
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z 
All Classes and Interfaces|All Packages|Constant Field Values|Serialized Form