Classes¶

• namespace nb::literals
• namespace std
• namespace tenzr
  • class tenzr::AddNode
    Elementwise addition of two operands.
  • class tenzr::ArgSortNode
  • class tenzr::Array
  • struct tenzr::CachedExecutable
    A compiled PJRT executable together with its hash and the shape and dtype descriptors of its inputs and outputs.
  • class tenzr::ConcatenateNode
    Join multiple operands along an axis.
  • class tenzr::ConvertNode
    Convert an operand to another element type.
  • class tenzr::ConvolutionNode
  • class tenzr::CustomOpNode
    Node backed by a custom operator registered in the OpRegistry.
  • struct tenzr::DTypeTraits
  • struct tenzr::DTypeTraits< bfloat16 >
    DTypeTraits specialization mapping bfloat16 to DType::kBFloat16.
  • struct tenzr::DTypeTraits< bool >
    DTypeTraits specialization mapping bool to DType::kBool.
  • struct tenzr::DTypeTraits< double >
    DTypeTraits specialization mapping double to DType::kFloat64.
  • struct tenzr::DTypeTraits< float >
    DTypeTraits specialization mapping float to DType::kFloat32.
  • struct tenzr::DTypeTraits< float16 >
    DTypeTraits specialization mapping float16 to DType::kFloat16.
  • struct tenzr::DTypeTraits< float4_e2m1fn >
  • struct tenzr::DTypeTraits< float6_e2m3fn >
  • struct tenzr::DTypeTraits< float6_e3m2fn >
  • struct tenzr::DTypeTraits< float8_e3m4 >
    DTypeTraits specialization mapping float8_e3m4 to DType::kFloat8E3M4.
  • struct tenzr::DTypeTraits< float8_e4m3 >
    DTypeTraits specialization mapping float8_e4m3 to DType::kFloat8E4M3.
  • struct tenzr::DTypeTraits< float8_e4m3b11fnuz >
  • struct tenzr::DTypeTraits< float8_e4m3fn >
    DTypeTraits specialization mapping float8_e4m3fn to DType::kFloat8E4M3FN.
  • struct tenzr::DTypeTraits< float8_e4m3fnuz >
  • struct tenzr::DTypeTraits< float8_e5m2 >
    DTypeTraits specialization mapping float8_e5m2 to DType::kFloat8E5M2.
  • struct tenzr::DTypeTraits< float8_e5m2fnuz >
  • struct tenzr::DTypeTraits< float8_e8m0fnu >
  • struct tenzr::DTypeTraits< int1 >
    DTypeTraits specialization mapping int1 to DType::kInt1.
  • struct tenzr::DTypeTraits< int16_t >
    DTypeTraits specialization mapping int16_t to DType::kInt16.
  • struct tenzr::DTypeTraits< int2 >
    DTypeTraits specialization mapping int2 to DType::kInt2.
  • struct tenzr::DTypeTraits< int32_t >
    DTypeTraits specialization mapping int32_t to DType::kInt32.
  • struct tenzr::DTypeTraits< int4 >
    DTypeTraits specialization mapping int4 to DType::kInt4.
  • struct tenzr::DTypeTraits< int64_t >
    DTypeTraits specialization mapping int64_t to DType::kInt64.
  • struct tenzr::DTypeTraits< int8_t >
    DTypeTraits specialization mapping int8_t to DType::kInt8.
  • struct tenzr::DTypeTraits< uint1 >
    DTypeTraits specialization mapping uint1 to DType::kUInt1.
  • struct tenzr::DTypeTraits< uint16_t >
    DTypeTraits specialization mapping uint16_t to DType::kUInt16.
  • struct tenzr::DTypeTraits< uint2 >
    DTypeTraits specialization mapping uint2 to DType::kUInt2.
  • struct tenzr::DTypeTraits< uint32_t >
    DTypeTraits specialization mapping uint32_t to DType::kUInt32.
  • struct tenzr::DTypeTraits< uint4 >
    DTypeTraits specialization mapping uint4 to DType::kUInt4.
  • struct tenzr::DTypeTraits< uint64_t >
    DTypeTraits specialization mapping uint64_t to DType::kUInt64.
  • struct tenzr::DTypeTraits< uint8_t >
    DTypeTraits specialization mapping uint8_t to DType::kUInt8.
  • struct tenzr::DeviceInfo
    Description of one PJRT device: its ids, platform kind, debug string, and handle.
  • class tenzr::DivNode
    Elementwise division of two operands.
  • class tenzr::DynamicSliceNode
  • class tenzr::DynamicUpdateSliceNode
  • struct tenzr::ExecutableInfo
    Best-effort metadata for a compiled PJRT executable (fields may be unset by some plugins).
  • class tenzr::ExecutionEngine
    Compiles and executes tensor graphs, dispatching work through the hardware manager and JIT cache.
  • struct tenzr::ExecutionResult
    Result of executing a graph, holding device-resident buffers plus lazily materialized host storage for its outputs.
  • class tenzr::GatherNode
    Gather slices of an operand at the given indices.
  • class tenzr::HardwareManager
    Owns the loaded PJRT plugin and client, exposing device enumeration and platform metadata.
    • struct tenzr::HardwareManager::DeviceMemoryStats
      Device allocator statistics (from PJRT_Device_MemoryStats when the plugin supports it).
  • class tenzr::JitCache
    Process-wide cache of compiled executables keyed by namespaced structural hashes.
  • class tenzr::MatmulNode
    Matrix multiplication of two operands.
  • class tenzr::MemoryPool
    Process-wide cache of reusable PinnedBuffer allocations, keyed by size.
  • class tenzr::Mesh
    Named grid of devices defining axes used by sharding specifications.
  • class tenzr::MulNode
    Elementwise multiplication of two operands.
  • class tenzr::NamedSharding
    Pairs a device Mesh with a PartitionSpec to describe a tensor layout.
  • struct tenzr::NetworkParameters
    Latency and bandwidth constants of the modeled network.
  • class tenzr::Node
  • struct tenzr::OpDef
  • class tenzr::OpRegistrationBuilder
  • class tenzr::OpRegistry
    Process-wide registry of custom operator definitions, keyed by name.
  • class tenzr::PadNode
    Pad an operand with a fill value at the edges and between elements.
  • class tenzr::PartitionSpec
    Per-dimension list of mesh axis names describing how a tensor is partitioned.
  • class tenzr::PinnedBuffer
  • class tenzr::PjrtProfiler
    Drives the optional PJRT profiler extension and captures a serialized XPlane profile.
  • class tenzr::ReduceNode
  • class tenzr::ReduceWindowNode
  • class tenzr::ReluNode
    Elementwise rectified linear unit.
  • class tenzr::ReshapeNode
    View of an operand reshaped to a new shape.
  • class tenzr::ScatterNode
  • class tenzr::Shape
    Shape of an array: the ordered extents of its dimensions.
  • class tenzr::ShardingSpec
  • class tenzr::SigmoidNode
    Elementwise logistic sigmoid.
  • struct tenzr::Slice
    Per-axis slicing bounds (start, stop, step) used to restrict arrays and layouts.
  • class tenzr::SliceNode
    View of an operand restricted by per-axis slice bounds.
  • class tenzr::StridedLayout
  • class tenzr::SubNode
    Elementwise subtraction of two operands.
  • class tenzr::TenzrException
    Exception type produced by TENZR_THROW and TENZR_CHECK, carrying a formatted message.
  • class tenzr::TorusCostModel
    Analytic cost model comparing torus and ring all-reduce times.
  • struct tenzr::TorusDimensions
    Extents of a three-dimensional torus network.
  • class tenzr::TransposeNode
    View of an operand with its dimensions reordered by a permutation.
  • class tenzr::UFunc
  • class tenzr::UFuncRegistry
    Process-wide registry of named ufuncs, including the standard built-ins.
  • class tenzr::VariableNode
  • namespace tenzr::apiutils
  • namespace tenzr::autograd
  • struct tenzr::bfloat16
    16-bit brain floating-point element type stored as raw bits.
  • namespace tenzr::check
    • struct tenzr::check::DeviceTestTarget
      A test device target pairing its name with whether it is available on this machine.
    • class tenzr::check::HardwareHarness
      Test helper that discovers available device targets and runs a test body once per target.
  • namespace tenzr::checkpoint
    • class tenzr::checkpoint::CheckpointManager
      Saves and restores named state with retention and best-metric tracking.
    • struct tenzr::checkpoint::CheckpointMetadata
      Bookkeeping record for one saved checkpoint.
    • struct tenzr::checkpoint::CheckpointOptions
      Configuration for a CheckpointManager.
  • namespace tenzr::concepts
    Compile-time element-type constraints over DTypeTraits.
    • namespace tenzr::concepts::detail
  • namespace tenzr::data
    • class tenzr::data::BatchSampler
      Groups indices produced by a Sampler into fixed-size batches.
    • class tenzr::data::DataLoader
      Iterates a Dataset in batches with optional collation.
      • class tenzr::data::DataLoader::Iterator
        Forward iterator over the batches of a DataLoader.
    • struct tenzr::data::DataLoaderOptions
      Configuration for a DataLoader.
    • class tenzr::data::Dataset
      Abstract indexed collection of tensor samples.
    • class tenzr::data::RandomSampler
      Sampler that yields a shuffled permutation of the dataset indices.
    • class tenzr::data::Sampler
      Abstract source of index orders over a dataset.
    • class tenzr::data::SequentialSampler
      Sampler that yields indices in ascending order.
    • class tenzr::data::ShardedDataset
      View of a Dataset restricted to the samples owned by one shard.
    • class tenzr::data::TensorDataset
      Dataset backed by an in-memory list of tensors.
  • namespace tenzr::debug
  • namespace tenzr::detail
  • namespace tenzr::execution
    • class tenzr::execution::Communicator
      Process-group communication handle exposing its rank, its size, and a barrier.
    • class tenzr::execution::DeviceMesh
      Logical grid of devices with optional named dimensions, mapping between device ids and mesh coordinates.
    • class tenzr::execution::ExecutionStream
      Serialized work queue backed by a worker thread, used to submit and synchronize device work.
    • class tenzr::execution::StreamEvent
      Handle to a recorded point in a stream's work, used to query completion or to make later work wait.
    • struct tenzr::execution::StreamEventState
      Shared readiness state for a stream event, holding the ready flag, the PJRT event, and completion callbacks.
  • struct tenzr::float16
  • namespace tenzr::io
    • class tenzr::SafetensorsFile
      Read-only memory-mapped view of a safetensors file exposing its tensors, descriptors, and metadata.
    • struct tenzr::TensorDescriptor
      Header metadata for one tensor in a safetensors file: name, dtype, shape, and payload byte range.
  • namespace tenzr::lax
  • namespace tenzr::modules
  • namespace tenzr::nn
    • class tenzr::nn::BCELoss
      Callable wrapper around the binary cross-entropy loss.
    • class tenzr::nn::BatchNorm
      Batch normalization module over the channel dimension.
    • class tenzr::nn::Conv2D
      2-D convolution module with configurable kernel, stride, and padding.
    • class tenzr::nn::CrossEntropyLoss
      Callable wrapper around the cross-entropy loss over logits.
    • class tenzr::nn::Dropout
      Dropout regularization module active only in training mode.
    • class tenzr::nn::Embedding
      Lookup-table module mapping integer indices to dense vectors.
    • class tenzr::nn::GELU
      Gaussian error linear unit activation module.
    • class tenzr::nn::LayerNorm
      Layer normalization module over the last dimension.
    • class tenzr::nn::Linear
      Fully connected affine module applying a weight and optional bias.
    • class tenzr::nn::LogSoftmax
      Log-softmax module normalizing over a configurable axis.
    • class tenzr::nn::MSELoss
    • class tenzr::nn::Module
      Base class for neural-network layers holding parameters, submodules, and training state.
    • class tenzr::nn::RMSNorm
      Root-mean-square normalization module.
    • class tenzr::nn::ReLU
      Rectified linear unit activation module.
    • class tenzr::nn::Sequential
      Container module that applies its layers in order.
    • class tenzr::nn::SiLU
      SiLU (swish) activation module.
    • class tenzr::nn::Sigmoid
      Logistic sigmoid activation module.
    • class tenzr::nn::Softmax
      Softmax module normalizing over a configurable axis.
    • class tenzr::nn::Tanh
      Hyperbolic tangent activation module.
  • namespace tenzr::numpy
    • struct tenzr::numpy::BroadcastResult
      Pair of arrays broadcast to a common shape.
  • namespace tenzr::optim
    • class tenzr::optim::Adam
      Adam optimizer with optional AMSGrad and decoupled weight decay.
    • class tenzr::optim::AdamW
      Adam variant that applies decoupled weight decay.
    • class tenzr::optim::ExponentialMovingAverage
      Maintains an exponential moving average of a parameter set.
    • class tenzr::optim::GradientTransformation
      Abstract gradient transformation defining Init and Update for an optimizer.
    • class tenzr::optim::Lion
      Lion optimizer using sign-based updates with first- and second-moment decays.
    • struct tenzr::optim::OptState
      Optimizer state holding the step counter and per-parameter tensors.
    • class tenzr::optim::Optimizer
      Binds a parameter set to a gradient transformation and applies update steps.
    • class tenzr::optim::SGD
      Stochastic gradient descent with momentum, weight decay, and Nesterov options.
  • namespace tenzr::plax
    • struct tenzr::plax::TopKResult
      Values and indices produced by a top-k reduction.
  • namespace tenzr::profiler
    • struct tenzr::profiler::BottleneckReport
      Heuristic analysis of the dominant performance bottleneck.
    • struct tenzr::profiler::ProfileSummary
      Aggregated profile statistics for a recorded trace.
    • class tenzr::profiler::Profiler
      Process-wide collector of profile events with Chrome Trace export.
    • class tenzr::profiler::ScopedTrace
      RAII helper that records a duration for its enclosing scope.
    • class tenzr::profiler::StepMarker
      RAII helper that marks the start and end of a training step.
    • struct tenzr::profiler::TopOp
      Aggregated timing statistics for one operation in a profile.
    • struct tenzr::profiler::TraceEvent
      One Chrome Trace Event entry recorded by the profiler.
    • class tenzr::profiler::TraceSession
      RAII helper that starts profiling on construction and exports on destruction.
  • namespace tenzr::python
    • namespace tenzr::python::dlpack_abi
      • struct tenzr::python::dlpack_abi::DLDataType
      • struct tenzr::python::dlpack_abi::DLDevice
      • struct tenzr::python::dlpack_abi::DLManagedTensor
      • struct tenzr::python::dlpack_abi::DLManagedTensorVersioned
      • struct tenzr::python::dlpack_abi::DLPackVersion
      • struct tenzr::python::dlpack_abi::DLTensor
  • namespace tenzr::random
    • class tenzr::random::Generator
      Mutex-protected pseudo-random number generator built on a 64-bit Mersenne twister.
    • struct tenzr::random::PRNGKey
      Functional pseudo-random key formed by two 32-bit words.
    • struct tenzr::random::SplitKeys
      Two independently derived PRNG keys.
  • namespace tenzr::runtime
    • class tenzr::runtime::DeviceBuffer
      RAII owner of a PJRT device buffer that awaits in-flight work before releasing it.
  • namespace tenzr::scipy
    • namespace tenzr::scipy::linalg
    • namespace tenzr::scipy::optimize
      • struct tenzr::scipy::optimize::OptimizeResult
        Outcome of a scipy.optimize minimization routine.
    • namespace tenzr::scipy::special
  • namespace tenzr::sharding
    • struct tenzr::sharding::ShardingAnalysis
      Derived Shardy and GSPMD representations of a NamedSharding.
  • namespace tenzr::txpr
    • struct tenzr::txpr::AttributePayload
      Attribute values attached to a Txpr equation, covering layout, padding, reduction, sorting, and collective parameters.
    • class tenzr::txpr::CustomVJPFunction
      Callable wrapper that applies a function decorated with a registered custom VJP rule.
    • class tenzr::txpr::CustomVJPRegistry
      Process-wide registry mapping operation names to custom VJP rules.
    • struct tenzr::txpr::CustomVJPRule
      A forward/backward rule pair defining the gradient of an operation that autodiff cannot derive on its own.
    • class tenzr::txpr::JittedFunction
      Callable wrapper produced by Jit that traces, compiles, and caches the wrapped function on first use.
    • struct tenzr::txpr::JvpResult
      Primal output and its tangent produced by a forward-mode JVP.
    • struct tenzr::txpr::LinearizeResult
      Primal output and the linearized JVP function produced by Linearize.
    • struct tenzr::txpr::ScanResult
      Carry and stacked outputs produced by a scan.
    • class tenzr::txpr::TracerContext
      Per-thread stack of active tracing targets; traced operations append to the innermost active Txpr.
    • class tenzr::txpr::TracerScope
      RAII guard that pushes a Txpr as the active tracing target and pops it on destruction.
    • class tenzr::txpr::Txpr
      SSA-style intermediate representation of a computation as typed variables and equations; produced by tracing and consumed by emitters and validators.
    • struct tenzr::txpr::TxprCompiledModule
      A lowered and compiled Txpr module pairing the emitted MLIR text with its input and output variable descriptors.
    • class tenzr::txpr::TxprEmitter
      Lowers a Txpr into StableHLO MLIR text by assigning each variable a unique SSA name.
    • struct tenzr::txpr::TxprEqn
      A single Txpr equation: an opcode, its input and output variables, and its attributes.
    • struct tenzr::txpr::TxprVar
      A Txpr variable reference pairing a VarId with its shape and dtype.
    • struct tenzr::txpr::ValueAndGradResult
      Value and gradient produced by a value-and-grad transform.
    • struct tenzr::txpr::VarDef
      Type and shape definition of a Txpr variable, including its optional debug name.
    • struct tenzr::txpr::VarId
      Typed identifier for a variable within a Txpr, wrapping a uint32 index; the sentinel value marks an invalid id.
    • struct tenzr::txpr::VjpResult
      Primal output and pullback produced by a multi-argument VJP.
    • class tenzr::txpr::VmappedFunction
      Callable wrapper produced by Vmap that traces the wrapped function unbatched and vectorizes it over a batch dimension.
    • namespace tenzr::txpr::detail
  • namespace tenzr::ufunc
• namespace tenzr
• namespace tenzr
• namespace tenzr
• namespace tenzr
• namespace tenzr
• namespace tenzr
• namespace tenzr
• namespace tenzr::numpy
• namespace tenzr::plax
• namespace tenzr::python
• namespace tenzr::python
• namespace tenzr::random
• namespace tenzr::txpr
• namespace tenzr::txpr
• namespace tenzr::txpr
• namespace tenzr::txpr
• namespace tenzr::txpr
• namespace tenzr::txpr