Index
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,epsand 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); withamsgrad, 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;
decoupledmakes 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;
decoupledmakes the weight decay AdamW's,amsgraduses 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,
epsand 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
graphastaskId. - 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
-
shortcutnull 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 ofF, 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), andProcessGroup, 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,TrainerandPredictorrun 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):
GgufFilegives 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:ModuleandSequential, 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:
OnnxSessionloads an ONNX file and runs it on jitNN's operators. - ai.beestream.jitnn.tf - package ai.beestream.jitnn.tf
-
The TensorFlow importer:
TfModelloads 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
tensorreplaced, 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 gradientgradOutof 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 examplellama, ornullwhen 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
srcinto 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
attentionrecord 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'sis_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, orfallbackwhen 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
Attributerecord class. - attributes() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
Returns the value of the
attributesrecord component. - attrInt(String, long) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
Integer attribute
key, orfallbackwhen the node does not set it. - attrInts(String, long[]) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
Integer-list attribute
key, orfallbackwhen the node does not set it. - attrString(String, String) - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
String attribute
key, orfallbackwhen 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 ofF.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
kernelwith 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
kernelxkernelwindows (stridekernel, no padding). - AvgPool2d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.AvgPool2d
-
Pools
kernelxkernelwindows 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
kernelwith 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
batchrecord component. - batch(int, int, float[], int[]) - Method in class ai.beestream.jitnn.data.Blobs
-
Copies batch
bofsizesamples intoxbandyb. - 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
bofsizesamples, in dataset order, intoxandy. - batch(int, int, float[], int[]) - Method in class ai.beestream.jitnn.data.Mnist
-
Copies batch
b(of sizesize) of images intoxand labels intoy. - Batch(int[], int[], int[], int, int) - Constructor for record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
-
Creates an instance of a
Batchrecord class. - Batch(Tensor, Tensor, int, int) - Constructor for record class ai.beestream.jitnn.data.DataLoader.Batch
-
Creates an instance of a
Batchrecord 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
channelschannels 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
hiddenunits,typeVocabtoken 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
vocabtokens and up tocontextpositions, withlayersblocks ofdimfeatures inheadsheads, initialised fromseed. - 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
pin [0, 1] againsttarget, 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 asreductionsays. - 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
-
F.binaryCrossEntropyWithLogits(Tensor, Tensor)reduced asreductionsays. - 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:
classesGaussian clusters indimsdimensions. - Blobs(int, int, int, float, long) - Constructor for class ai.beestream.jitnn.data.Blobs
-
countsamples aroundclassescentres drawn uniformly in [-2, 2)^dims, with Gaussian noise of standard deviationspread, fromseed. - 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
-
groupswork-groups oflocalthreads. - blocks2D(int, int, int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
-
gx * gywork-groups oflx * lythreads. - blocks3D(int, int, int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
-
gx * gy * gzwork-groups oflxthreads. - 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
tensorreplaced, on every rank, by rankroot'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
Bucketssizes, 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
sentencesseen at leastminCounttimes. - 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
datain batches ofbatchSize. - 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
bytesrecord component. - bytes() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
-
Returns the value of the
bytesrecord 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
elementsvalues (elementsa multiple of the block size).
C
- c() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
-
Returns the value of the
crecord 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
batchsize, positions padded at the end to aseqsize withpadToken, 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
vocabtokens: embeddings ofdim,layersLSTM (or GRU, withuseGru) layers ofhiddenfeatures, and the head. - check(String, Tensor) - Method in class ai.beestream.jitnn.io.StateDict
- checkpoint(Function, Tensor...) - Static method in class ai.beestream.jitnn.Grad
-
Activation checkpointing, as PyTorch's
torch.utils.checkpoint: runssegmentoninputsand 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
axisintochunksequal 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
maxNormat the nextOptimizer.step(), as PyTorch'sclip_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
paramsso that their global L2 norm is at mostmaxNorm, as PyTorch'sclip_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
keyas a float, orotherwisewhen absent. - configInt(String...) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
-
The first of
keyspresent in the config, as an int; throws when none is. - configString(String, String) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
-
Config value
keyas a string, orotherwisewhen 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
similar1 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:
padsholds 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
intooutchannels, initialised fromseed. - Conv1d(int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv1d
-
With
groupsgroups 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.Conv2dsemantics 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
intooutchannels ingroupsgroups (both divisible by it), initialised fromseed. - Conv2d(int, int, int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.Conv2d
-
A convolution from
intooutchannels with akernelxkernelkernel 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
intooutchannels with a cubic kernel, initialised fromseed. - Conv3d(int, int, int, int, int, int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Conv3d
-
With
groupsgroups 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.ConvTranspose2dsemantics 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
intooutchannels ingroupsgroups, initialised fromseed. - ConvTranspose2d(int, int, int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.ConvTranspose2d
-
A transposed convolution from
intooutchannels 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 tensorname, in place, so plans compiled withtargetkeep 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
sentencesof word ids, negatives drawn by the word countscounts(indexed by id), randomness fromseed. - 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
minLrovertMaxepochs (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 withjitnn.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 INT32labels[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
nodeto 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
nodeon 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
lengthentries of dimensionaxisoft, 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
datarecord component. - DataLoader - Class in ai.beestream.jitnn.data
-
Batches of a
Datasetfor training, as PyTorch'sDataLoader: 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 replicamodel; ranks synchronise throughcollective. - 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
dataTyperecord 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; withskipSpecial, 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
optiona 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
dimNamesrecord 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
dimsrecord component. - dims() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
-
Returns the value of the
dimsrecord 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
widthchannels in its first layer, initialised fromseed. - 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
documentsdocuments overvocabwords, vectors ofdim, trained asmodesays, initialised fromseed. - 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
domainrecord 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:
seedis 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
pand scales the rest by 1 / (1 - p); masks are seeded fromseed. - dtype() - Method in enum class ai.beestream.jitnn.gguf.GgmlType
-
The jitNN type this one loads as, or
nullwhen jitNN has none. - dtype() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
-
Returns the value of the
dtyperecord component. - dtype() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
-
Returns the value of the
dtyperecord 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
elemTyperecord 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
alphafor 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 INT32index: shape index.shape + [D]. - embedding(Tensor, Tensor, int) - Static method in class ai.beestream.jitnn.F
-
As
F.embedding(Tensor, Tensor); rowpaddingIdxreceives no gradient. - Embedding - Class in ai.beestream.jitnn.nn
-
A lookup table of
countvectors of sizedim, applied to INT32 indices. - Embedding(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.Embedding
-
Row
paddingIdxreceives no gradient (pass -1 for none). - Embedding(int, int, long) - Constructor for class ai.beestream.jitnn.nn.Embedding
-
A table of
countvectors of sizedim, standard normal fromseed, 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
weightpicked by the INT32indices, combined per bag bymode. - 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
countvectors of sizedim, standard normal fromseed, whose bags combine asmodesays. - emit(Emitter, Node) - Method in interface ai.beestream.jitnn.CustomOp.Launcher
-
Adds the tasks that compute
nodethroughe. - 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];
maskis 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(andbwhen not null) into at mostmaxLengthids, 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
- Encoding(int[], int[], String[]) - Constructor for record class ai.beestream.jitnn.nn.text.Tokenizer.Encoding
-
Creates an instance of a
Encodingrecord 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
entriesrecord component. - Entry(DType, String, long[], long, long) - Constructor for record class ai.beestream.jitnn.io.SafeTensors.Entry
-
Creates an instance of a
Entryrecord 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
seqtokens. - 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
targetshape. - 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
-
OnnxExport.export(Path, Function, Tensor[], String[], String)for a one-input model. - export(Path, Function, Tensor[], String[], String) - Static method in class ai.beestream.jitnn.onnx.OnnxExport
-
Records
forwardoninputs(put the model in evaluation mode first) and writes it tofilewith graph inputsinputNamesand the outputoutputName. - 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 tensorlike(ONNX EyeLike).
F
- f() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
-
Returns the value of the
frecord component. - F - Class in ai.beestream.jitnn
-
Functional operators that are not methods of
Tensor. - F.BagMode - Enum Class in ai.beestream.jitnn
-
How
F.embeddingBag(Tensor, int[], Tensor, F.BagMode)andF.segmentReduce(Tensor, int[], F.BagMode)combine the rows of a bag. - 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
-
How
F.pool(Tensor, F.PoolMode, int[], int[], int[], int[], boolean, boolean, float)combines a window. - 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
factorrecord component. - factor() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
-
Returns the value of the
factorrecord 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,lowestbeing 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
dimto 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
-
The results of
Collective.startAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int)(same tensors, buffer and tag), asCollective.allReduced(int, Tensor[], Collective.ReduceOp)returns them. - finishAllReduce(int, Tensor[], Tensor, Collective.ReduceOp, int) - Method in class ai.beestream.jitnn.dist.NcclCollective
-
Waits for the all-reduce started with
tagand 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
floatsrecord 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
nameas 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
nameas a float; throws when it is missing. - floatValue(String, float) - Method in class ai.beestream.jitnn.gguf.GgufFile
-
GgufFile.floatValue(String), orfallbackwhen 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;
Sequentialuses 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-uncasedand other BERT checkpoints, asBertModel,BertForMaskedLMorBertForPreTraining);roberta(roberta-baseand 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.jsonand safetensors weights, asGPT2ModelorGPT2LMHeadModelsave 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, ornullfor 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
grecord component. - gather(int, Tensor) - Method in class ai.beestream.jitnn.Tensor
-
torch.gather along
axiswith 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
latentvalues,widthchannels in its last hidden layer, initialised fromseed. - 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, ornullwhen 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 jitNNDTypeeach 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
GgufTensorInforecord 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
vocabwords with vectors ofdimand weighting f(x) = min(1, (x / xMax)^alpha), initialised fromseed. - 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
vocabtokens and up tocontextpositions, withlayersblocks ofdimfeatures inheadsheads, initialised fromseed. - 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:
roundstrees of depth at mostmaxDepth, shrunk bylearningRate, with at leastminSamplesLeafsamples per leaf, L2 penaltylambdaon the leaf values, and features in at mostmaxBinsbins. - 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. afterloss.item()) halves the scale on an overflow, or doubles it aftergrowthIntervalclean 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
initScalethat multiplies the scale bygrowthFactoraftergrowthIntervalsteps without overflow and bybackoffFactorafter 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
Graphrecord 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
intooutfeatures: a linear map without bias initialised fromseed, and a bias of zeros. - grid1D(int) - Static method in class ai.beestream.jitnn.engine.Emitter
-
A 1-D grid of at least
nthreads, in groups ofEmitter.LOCAL. - grid2D(int, int, int) - Static method in class ai.beestream.jitnn.engine.Emitter
-
A 2-D grid covering
x * ythreads in tiles oftile * tile. - groupNorm(Tensor, int, Tensor, Tensor, float) - Static method in class ai.beestream.jitnn.F
-
GroupNorm (Wu and He) of x [N, C, ...] in
groupsgroups 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
channelschannels ingroupsgroups, with the given eps, with or without the per-channel affine. - groups(int) - Static method in class ai.beestream.jitnn.engine.Emitter
-
groupswork-groups ofEmitter.LOCALthreads 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
-
As
F.gru(Tensor, Tensor, Tensor, Tensor, Tensor, Tensor)for padded batches oflengths. - GRU - Class in ai.beestream.jitnn.nn
-
A stack of GRU layers over [T, B, I] sequences (time first), optionally bidirectional, PyTorch's
nn.GRUequations, 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
-
layersGRU layers frominputtohiddenfeatures, optionally bidirectional, initialised fromseed. - GRU(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.GRU
-
layersunidirectional GRU layers frominputtohiddenfeatures, initialised fromseed. - 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
hrecord component. - h() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
-
Returns the value of the
hrecord 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
Headerrecord class. - HfCheckpoint - Class in ai.beestream.jitnn.nn.models
-
A Hugging Face model directory:
config.jsonand the weights inmodel.safetensorsor in shards listed bymodel.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
highFreqFactorrecord 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
worldSizeranks 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
irecord 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
idArrayrecord 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
idsrecord 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
indexrecord 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
initializersrecord 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
runtakes 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
inputsrecord component. - inputs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
Returns the value of the
inputsrecord 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
configthe 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
channelschannels 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), orfallbackwhen 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
intsrecord 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
nameas 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
dirholds the MNIST training images (plain or gzipped). - isSet(JitnnConfig.Option) - Method in class ai.beestream.jitnn.JitnnConfig
-
Whether
optionwas 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.loadreads ONNX files through jitnn-onnx. - JitnnEngineProvider - Class in ai.beestream.jitnn.djl
-
Registers
JitnnEnginewith 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 bycause. - 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, arraysLists, integersLong, other numbersDouble.
K
- keep(Tensor) - Static method in class ai.beestream.jitnn.Grad
-
Marks
tas 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),
nInitrestarts 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
kclusters, at mostmaxIteriterations per run, tolerancetol,nInitruns, seeding fromseed. - 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
knearest 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,
epsand 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
etaandeps. - 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
featuresvalues, with eps 1e-5 and a bias. - LayerNorm(int, float, boolean) - Constructor for class ai.beestream.jitnn.nn.LayerNorm
-
LayerNorm over
featuresvalues 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
lengthrecord 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 samekey. - libraryJoin(Native.Lib, String, Emitter.TaskAdder, String) - Method in class ai.beestream.jitnn.engine.Emitter
-
The task that joins the side work
Emitter.libraryForked(Native.Lib, String, Emitter.TaskAdder, String, Emitter.TaskAdder)started withkey. - 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
Linearrecord class. - Linear(int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.Linear
-
A linear layer from
intooutfeatures, with or without a bias, initialised fromseed. - Linear(int, int, long) - Constructor for class ai.beestream.jitnn.nn.Linear
-
A linear layer from
intooutfeatures with a bias, initialised fromseed. - 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
cand at mostmaxIteriterations. - 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
vocabtokens withlayersblocks ofdimfeatures inheadsheads, a feed-forward ofhiddenfeatures and rotary baseropeBase, initialised fromseed. - Llama(int, int, int, int, int, int, float, float, boolean, boolean, long) - Constructor for class ai.beestream.jitnn.nn.models.Llama
-
A Llama with
kvHeadskey/value heads (grouped-query attention when fewer thanheads), RMSNorm epsiloneps, the output projection tied to the token embedding whentied, and biases on the q, k and v projections whenqkvBias. - 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
Llama3record class. - load(Tensor, String) - Method in class ai.beestream.jitnn.nn.models.HfCheckpoint
-
Copies weight
nameinto parametertargetof the same shape. - load(Path) - Static method in class ai.beestream.jitnn.data.TextCorpus
-
The bytes of
fileas 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 directorypath. - 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_defaultsignature, or a frozen GraphDef file. - load(Path, boolean) - Static method in class ai.beestream.jitnn.data.Cifar10
-
diris 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
signatureof 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
namesintotarget. - loadScalars(StateDict) - Method in class ai.beestream.jitnn.nn.optim.NAdam
- loadScalars(StateDict) - Method in class ai.beestream.jitnn.nn.optim.Optimizer
-
Restores what
Optimizer.saveScalars(StateDict)saved. - 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
nameintotarget(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
sizeneighbouring 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
cand at mostmaxIterL-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
logvarrecord 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
nameas 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
lowFreqFactorrecord 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: INT32lengths[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.LSTMequations, 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
-
layersLSTM layers frominputtohiddenfeatures, optionally bidirectional, initialised fromseed. - LSTM(int, int, int, long) - Constructor for class ai.beestream.jitnn.nn.LSTM
-
layersunidirectional LSTM layers frominputtohiddenfeatures, initialised fromseed. - 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.mdshows 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):
modeltakes tokens [B, T] and a mask [B, 1, 1, T] or [B, 1, T, T] (masknull 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
matrixrecord 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 whenkeepDimis 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
kernelwith 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
kernelwith 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
kernelwith 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
kernelxkernelwindows (stridekernel, no padding). - MaxPool2d(int, int, int) - Constructor for class ai.beestream.jitnn.nn.MaxPool2d
-
Pools
kernelxkernelwindows 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
kernelwith 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 whenkeepDimis 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
metadatarecord 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 whenkeepDimis 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
inChannelschannels withclassesoutputs and dropoutdropoutbefore the classifier, initialised fromseed. - 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
inputandtarget(same shape). - mseLoss(Tensor, Tensor, F.Reduction) - Static method in class ai.beestream.jitnn.F
-
Squared error between
inputandtarget(same shape), reduced asreductionsays. - 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
murecord 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.MultiheadAttentionwith batch_first: q, k, v = split(x Wqkv^T + bqkv),headsheads, then the output projection. - MultiheadAttention - Class in ai.beestream.jitnn.nn
-
Multi-head self-attention on [B, T, D] inputs, like PyTorch's
nn.MultiheadAttentionwith batch_first: one fused q/k/v projection [3D, D],headsheads of D / heads features, and an output projection. - MultiheadAttention(int, int, boolean, long) - Constructor for class ai.beestream.jitnn.nn.MultiheadAttention
-
Attention over
dimfeatures inheadsheads (dimdivisible by it), with or without projection biases, initialised fromseed. - 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
stepup to and includingmax(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
namerecord component. - name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Attribute
-
Returns the value of the
namerecord component. - name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
-
Returns the value of the
namerecord component. - name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
Returns the value of the
namerecord component. - name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.TensorData
-
Returns the value of the
namerecord component. - name() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.ValueInfo
-
Returns the value of the
namerecord 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 toposInfand -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 - 1alongaxis, 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 fortornado --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 ofworldSize. - 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
-
nwords 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
labelsunder log-probabilitieslogp. - 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
Noderecord class. - nodes() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Graph
-
Returns the value of the
nodesrecord 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
sizenon-zero elements, the rest of the rowsfill. - 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
datain the shapedims(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:
xholds the samples back to back,yone 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));halfPrecisionas inOnnxSession.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
offsetrecord component. - offset() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
-
Returns the value of the
offsetrecord 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
bodywith this thread's plans on devicespec("driver:device", asjitnn.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
hookwith 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
paramswith learning ratelr. - 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
opTyperecord component. - originalContext() - Method in record class ai.beestream.jitnn.RopeScaling.Llama3
-
Returns the value of the
originalContextrecord 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
Outputrecord class. - Output(Tensor, Tensor, Tensor) - Constructor for record class ai.beestream.jitnn.nn.LSTM.Output
-
Creates an instance of a
Outputrecord class. - Output(Tensor, Tensor, Tensor) - Constructor for record class ai.beestream.jitnn.nn.models.Vae.Output
-
Creates an instance of a
Outputrecord 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
outputsrecord component. - outputs() - Method in record class ai.beestream.jitnn.onnx.OnnxModel.Node
-
Returns the value of the
outputsrecord 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, andDataParallel.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]andafter[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
-
twith dimensionaxisgrown tosize, the new elements set tovalue; 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]andafter[k]elements;valuefills 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
componentscomponents. - 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:
padsholds the padding before every axis, then after every axis;ceilrounds the output size up (a window must still start inside the input or its leading padding); average pooling counts the padding whencountIncludePad(never the positions added byceil); 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
minup to and includingmax(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
fitonly). - 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
weightof 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
parametersslopes (1 shared, or one per channel), eachinitat 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 oneCollective. - 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
partunderprefix.. - 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
padzeros 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
treesbootstrapped trees of depth at mostmaxDepth, 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
rawrecord 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 fromfileintotarget, 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
reconrecord 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
axisremoved (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
plansmore compiled plans in the plan cache (per concurrency slot), beyondjitnn.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
featuresvalues 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, withcentered, 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
-
Llama 3.1's rescaling; see
RopeScaling.llama3(double, double, double, int). - 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
-
modeloverx[N, ...] with N padded to abatchsize 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]
seedtensor (seeTensor.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
nodefrom 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
srecord 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 withsafetensors.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
iintoxatxAtand its label intoyatyAt(DataLoaderassembles 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
pathas 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
srcwritten atindex(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
srcadded atindexalongaxis. - 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
nesterovis set. - shape() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
-
Returns the value of the
shaperecord component. - shape() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
-
Returns the value of the
shaperecord component. - shape() - Method in record class ai.beestream.jitnn.io.SafeTensors.Source
-
Returns the value of the
shaperecord 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
numelelements 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
-
Emitter.sharedScratch(String, int)for FP16 buffers. - 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
seedand 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
sizerecord 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 ifnis 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 - 1alongaxis; 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
upperor lower triangle is read, and withunitDiagonalnot 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
Sourcerecord 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
axisinto 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 persistentbuffer(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
tensorsinto persistentbufferand starts its all-reduce on NCCL's side stream (seeCollective.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 withStateDict.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
putand 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 whenkeepDimis 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
targetshape (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
wantedto 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
rankof anNcclCollective; 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
trecord 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;adderadds 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
nameas a persistent jitNN tensor, memory-mapped where possible (see the class comment). - tensor(String) - Method in class ai.beestream.jitnn.io.StateDict
-
Tensor
nameas a new persistent tensor (a weight that stays on the device). - tensor(String, boolean) - Method in class ai.beestream.jitnn.gguf.GgufFile
-
Tensor
name;copyreads 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
TensorDatarecord 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(lengthTensor.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
textwithout 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
tokensrecord 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=valueproperties. - 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
rof batchb: 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
shapeatx[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
- 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
componentsdimensions 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
typerecord 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
typerecord component. - typeId() - Method in record class ai.beestream.jitnn.gguf.GgufTensorInfo
-
Returns the value of the
typeIdrecord component. - typeIdArray() - Method in record class ai.beestream.jitnn.nn.text.Tokenizer.Batch
-
Returns the value of the
typeIdArrayrecord 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
typeIdsrecord component. - typeName() - Method in record class ai.beestream.jitnn.io.SafeTensors.Entry
-
Returns the value of the
typeNamerecord 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
inchannels,widthchannels at the first level (doubling at each level down), initialised fromseed. - 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 (seeTensor.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
tafter 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
optdivide 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
loadcall 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;alignCornersas 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
infeatures throughhiddento a latent oflatentvalues, initialised fromseed. - 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
Valuedrecord 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
ValueInforecord 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
valuesrecord 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
minCounttimes.
W
- warmupCosine(Optimizer, int, int, float) - Static method in class ai.beestream.jitnn.nn.optim.LRScheduler
-
Linear warm-up over
warmupsteps, then a half-cosine decay tominRatiotimes the base rate attotalsteps (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
warmupsteps, then linear decay to 0 attotalsteps (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
bodyonbackend, syncs, and restores the previous backend. - with(JitnnConfig.Option, T) - Method in class ai.beestream.jitnn.JitnnConfig
-
A copy with
optionset tovalue(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
sizesthat 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
vocabwords with vectors ofdim, trained asmodesays, initialised fromseed. - 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
dataitself, without a copy, as PyTorch'sfrom_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
-
Tensor.wrap(float[], int...)for int data. - 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
-
eas aJitnnExceptionwhen it is a device or runtime failure (one from TornadoVM), elseeitself. - write(Path, Map, Map) - Static method in class ai.beestream.jitnn.io.SafeTensors
-
Writes
tensors, in iteration order, andmetadatatofile.
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
xrecord 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
yrecord component. - y() - Method in record class ai.beestream.jitnn.nn.GRU.Output
-
Returns the value of the
yrecord component. - y() - Method in record class ai.beestream.jitnn.nn.LSTM.Output
-
Returns the value of the
yrecord 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
All Classes and Interfaces|All Packages|Constant Field Values|Serialized Form