Class List

environment-neutral discrete actor/critic contract. Custom policies only need to implement evaluate; PPO collection, loss construction and updates do not depend on the concrete network architecture.

Class in the Ml module.

Decoupled weight decay — preferred for transformers.

AdaptiveAvgPool2d: Adaptive average pooling layer.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Semantic planar float32 audio value composed over oa::Matrix. The matrix view is [channels, samples]; sample rate and layout stay attached to that storage.

Class in the Audio module.

Class in the Audio module.

Class in the Audio module.

— per-tensor autograd state (PyTorch-style nullptr-optimized). Inference tensors pay zero bytes: autograd_ stays nullptr. Invariants: - leaf ⇔ !gradFn - leaf with requiresGrad owns a persistent grad matrix (Tier 1) - non-leaf has gradFn; grad is not populated unless retainGrad (v1: unsupported)

AvgPool2d: 2D average pooling layer.

Class in the Ui module.

BatchNorm2d: Batch normalization layer with running statistics.

Learns byte pair merges to build vocabulary from 256 base bytes to target size. Useful for compressing sequences and improving model efficiency.

BYTE EMBEDDING - The 256-entry lookup table Maps each byte value (0-255) to a d_model dimensional vector. input: [batch, seq_len] of UInt8 output: [batch, seq_len, d_model] of Float32 This replaces the massive 30k-100k token embedding tables in GPT/LLaMA. Ours is always 256 x d_model. Tiny. Fast. Universal.

BYTE ENCODER - Raw bytes to tensor and back

BYTE OUTPUT HEAD - convert hidden states back to byte probabilities input: [batch, seq_len, d_model] output: [batch, seq_len, 256] (logits over byte values)

base callback interface — all methods are optional (no-op by default)

Class in the Ui module.

Class in the Vision module.

handle-free liveness contract for one captured physical resource. resources that have not yet gained schema-v2 semantic values still appear here, which lets compatibility operations protect loss/readback outputs during migration.

Compact default MLP used by discrete-control PPO. It is a convenience model, not a restriction: callers can supply any ActorCritic implementation.

oa::CbCheckpoint keras modelCheckpoint(save_freq="epoch") + EarlyStopping's restore_best_weights, on top of oa::CheckpointManager. Every epoch end writes a resumable rotating checkpoint. Pass SaveEvery > 0 to add mid-epoch checkpoints every N completed optimizer steps. The master model is updated only on improvement. at each epoch end it prints a TF-style mini summary: epoch 3: cross_entropy improved from 0.4056 to 0.3486 — saving model epoch 4: cross_entropy did not improve from 0.3486 at train end, if RestoreBest is set and the best epoch wasn't the last one, the best checkpoint is loaded back into model + optimizer — you always walk away with the best weights, not whatever the final (possibly degraded) epoch produced. Monitored value: inMetric->result() when provided (e.g. a val_loss metric), otherwise the epoch mean train loss. Better/worse direction comes from the manager's lowerIsBetter config.

oa::CsvLoggerCallback Appends one exact row per completed optimizer step. Rate columns use explicit units; there is no logging cadence hidden inside the metric lifecycle.

Class in the Ml module.

oa::LrSchedulerCallback Applies any oa::LRScheduler to the optimizer at each step. Use oa::CosineScheduler, oa::OneCycleScheduler, oa::WarmupScheduler, etc. from <oa /ml/optim.h>.

oa::CbPhase Multi-phase training callback. phases are consecutive ranges of epochs — build the iterator with oa::ItTrainingConfig::epochSteps so epoch boundaries follow the phase schedule, then register one addPhase() per phase in order. Prints a schedule preview at train begin and a phase banner + phase-relative epoch headers (disable the progress bar's own header via SetShowEpochHeader): phase schedule: 1. warmup — 1 epoch (2000 steps) 2. main — 10 epochs (20000 steps) phase 1/2 — warmup · 1 epoch × 2000 steps epoch 1/1 2000/2000 |██████████| ... The OnPhaseBegin hook fires on entering each phase (including the first) — use it to swap LR schedulers, change datasets, etc.

oa::CbProgressBar tqdm/keras-hybrid progress bar with `█`+`░` and rolling metrics: epoch 1/5 938/938 |██████████| 0.65s · 0.7 ms/step · 1.26M sample/s · accuracy: 0.9091 · loss: 0.2914 epoch 2/5 938/938 |██████████| 0.59s · 0.6 ms/step · 1.31M sample/s · accuracy: 0.9134 · loss: 0.1822 mid-epoch: 468/938 |█████░░░░░| 0.32s · 0.7 ms/step · 1.25M sample/s · loss: 0.3128 Per-step updates rewrite the same line via ` `; epoch end leaves the final line and starts a new one. Uses lastLoss() so per- step refresh never forces a Sync. Accuracy comes from recordAccuracy(); if the caller hasn't set it (NaN), the field is omitted. Wall time per step and workload throughput are shown instead of GPU time. Latency and rates are derived from the iterator's single workload definition; they are not metrics that callers must register separately.

oa::CbSummary Prints a final training summary at onTrainEnd: loss, wall latency/throughput, GPU mean/p50/p95, the wall-to-GPU timing gap, and total duration. Optionally tracks initial loss for comparison. example output: summary: loss: initial 2.6558 · final 0.2965 · mean 0.4812 Wall: 0.07 ms/step · 943.68K sample/s · 60.40M token/s GPU: mean 0.051 ms/step · p50 0.049 · p95 0.061 · 1.26M sample/s run: 0.32s · 4686 steps · batch 64 · sequence 64 token/sample

training-specific callback base class. Subclass and attach via addCallback(). All hooks have default no-op implementations.

Class in the Ml module.

Class in the Core module.

Class in the Ml module.

CheckpointManager — auto-save best models with rotation. Wraps Module::save/load to manage directory structure, path naming, metric tracking, and incremental-checkpoint rotation.

Class in the Core module.

One quasi-simultaneous host/device clock sample. maximumDeviationNanoseconds is the vulkan implementation's uncertainty bound for the pair, not an OA estimate. Consumers should recalibrate long-running traces periodically.

Class in the Core module.

Class in the Core module.

Vulkan application — owns one factory-created, pinned oa::Engine around the tick loop. Subclass for training, inference, compute apps. Device selection: engineConfig_.devicePref / deviceIndex (see oa::EngineConfig).

Metadata row for docs, generators, and future validation (a stable type id + kind, like a node-type registry).

Class in the Core module.

Class in the Core module.

Class in the Core module.

Conv1d: 1D convolution layer (im2col + GEMM).

Conv2d: 2D convolution layer.

ConvTranspose1d: 1D transposed convolution (learnable upsampling); adjoint of Conv1d, no bias.

ConvTranspose2d: 2D transposed convolution layer (learnable upsampling).

Cosine annealing from maxLr to minLr over totalSteps.

SGDR: cosine annealing with periodic warm restarts. Period starts at t0 steps, multiplied by tMult after each restart.

Class in the Ml module.

Class in the Data module.

DataLoader (analogous to torch.utils.data.dataLoader)

(analogous to torch.utils.data.dataset)

— human-readable UTC date/time over the system clock. For logs, debugging, display. NEVER for consensus or deterministic math.

Class in the Vision module.

Class in the Ui module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Environment-neutral DQN update coordinator over caller-owned online and target modules, optimizer, and replay storage. It composes `ItTraining` for the exact optimizer lifecycle; it does not inherit a nominal trainer base.

module. training applies inverted dropout; evaluation is identity.

Class in the Data module.

Class in the Core module.

Embedding: Embedding lookup layer.

— empyrealm-style sequential modeling core. high-utilization reusable backbone: input projection (byte embed or custom) → mixer + flat per-token residual → mixed features [B*seq, dModel]. The mixer is EmpyrealmModule, which dispatches empyrealm* kernels (EmpyrealmDt, EmpyrealmAdt, EmpyrealmSiso) — renamed copies of the Mamba3* kernels with identical SPIR-V today, ready for future architecture-specific divergence. shader layout (for fusion / branding): Ssm/Mamba3/ — original Mamba3Siso* (untouched reference, used by Mamba3Module) Ssm/empyrealm/ — ported/copied starting point; will host fused empyrealm* variants (e.g. custom one-node mixers) while Mamba3 stays pristine. The reconstruction tutorial demonstrates the intended usage. Brand Mamba3 tech as empyrealm for custom evolution.

The engine is deliberately concrete and pinned. Optional presentation, video and collective services compose with it; they do not subclass it.

Class in the Core module.

Class in the Core module.

Stateful native environment session. The session borrows one Engine and privately owns its recorder/execution state. reset/step remain stateful commands; submit returns the exact completion for every command recorded since the previous completion. The public boundary never exposes context or queue controls.

Cheap non-owning completion value shared by compute, image, video, capture, rendering, and training. It is a GPU dependency first and a host wait handle second. queue submission consumes it through private Runtime lowering; public callers wait only at a deliberate CPU boundary.

move-only compiled executable work captured by oa::Engine. The plan borrows its engine and owns the graph and retained resources through exact completion. Destruction never waits: submitted work transfers to engine retirement.

Stable execution telemetry exposed after an explicit engine boundary. Mutable graph, recorder, and execution-session state remain private.

Class in the Ml module.

Class in the Ui module.

— stateless host-filesystem operations. oa::Path owns lexical path manipulation. oa::Paths owns OA location discovery.

Class in the Core module.

Flatten: Flatten layer with PyTorch-style start_dim/end_dim.

Class in the Ml module.

GPU sinusoidal embedding for normalized continuous time. Only the small, deterministic frequency vector is uploaded at construction; every batch is embedded by OA matrix kernels without a tensor-sized CPU loop or readback.

Reusable bidirectional Transformer backbone for flow/diffusion denoisers. input is conditioned token state [B,S,D] or flattened [B*S,D]; output keeps the same shape. This is a sibling model family that composes generic Transformer blocks; it is not part of the generic Transformer itself.

Class in the Core module.

Namespace in the Audio module.

Namespace in the Ml module.

Namespace in the Ui module.

Namespace in the Data module.

Namespace in the Vision module.

Namespace in the Ml module.

Namespace in the Crypto module.

Namespace in the Vision module.

Namespace in the Ml module.

Namespace in the Core module.

Namespace in the Ui module.

Namespace in the Ml module.

Namespace in the Ui module.

Namespace in the Core module.

Namespace in the Vision module.

Gelu: GELU activation layer.

Class in the Ui module.

oa::GradientTape — RAII scope guard that enables autograd for its lifetime. call backward() to execute the recorded backward graph.

oa::GradNo — RAII scope guard that disables autograd for its lifetime.

Class in the Core module.

Stable summary of one compiled executable graph. The graph and its mutable nodes remain private; training diagnostics expose only this value snapshot.

Class in the Ml module.

Class in the Ml module.

— 32-byte cryptographic hash (SHAKE-256 output)

Incremental hasher (SHAKE-256)

Hash-map/set hasher.

Class in the Core module.

Class in the Core module.

Identity: Identity passthrough layer.

— semantic wrapper around oa::Matrix

— uniform batch of images

Class in the Ui module.

RAII wrapper for a producer-owned DMA-BUF imported as a sampled VkImage. queue-family ownership starts at VK_QUEUE_FAMILY_FOREIGN_EXT; consumers must acquire/release it around GPU use before the producer reuses the buffer.

Class in the Ui module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Coordinates the two-phase lifecycle of synchronous on-policy training. Environment stepping, policy evaluation, and loss construction remain caller supplied; an owned `ItTraining` controls each exact optimizer update.

Class in the Ml module.

Class in the Crypto module.

LayerNorm: Layer normalization layer.

Linear: Fully connected linear layer.

Convenience scheduler: linear warmup to targetLr, then cosine annealing to minLr. Composes WarmupScheduler + CosineScheduler.

Class in the Core module.

A component is a small value, not a closed framework enum or a mutable registry entry. OA supplies neutral built-ins through Builtin; downstream libraries define their own named constexpr values from one-to-four-character tags without modifying OA or acquiring process-global state. namespace ChainLog { inline constexpr LogComponent Consensus{"CONS"}; }

Class in the Core module.

Class in the Core module.

Stateful log output is an explicitly owned host session. oa::Engine creates one from these options and selects it for the thread that owns the engine. The Log* macros use that selected session and fall back to stderr before an engine exists or on a thread that has not selected one.

(base)

— Mamba-3 selective state space model block Reference: Mamba-3 paper (https://arxiv.org/abs/2603.15569) Based on: https://github.com/state-spaces/mamba (mamba_ssm/modules/mamba3.py)

Read-only whole-file mapping with RAII ownership. Linux uses mmap. Other platforms retain a read-only owned byte buffer until a native mapping implementation is added.

A multidimensional semantic value backed by OA-managed storage. A matrix carries shape, element stride, dtype, device, byte-offset, and optional autograd metadata. Copying a matrix shares its storage and autograd state. Metadata-only views also alias storage; use `clone()` when independent storage is required. Matrix operations use OA's Vulkan execution path. Direct host observation is an explicit synchronization boundary, not a CPU compute fallback. Stateful device ownership remains with `oa::Engine`; a matrix only retains the storage required for its value and recorded graph lifetime.

MaxPool2d: 2D max pooling layer.

Class in the Core module.

memory capacity for value-only device descriptors. VkCpu reports host physical RAM because that identity is process-independent. Other vulkan descriptors do not identify a live allocator or physical device and return zero; query the exact oa::Engine for live vulkan budget/usage.

Class in the Crypto module.

Class in the Crypto module.

base class for all metrics. Follows the keras pattern: - update(): add a new batch of predictions/labels - reset(): clear accumulated state - result(): get the current metric value

Class in the Ml module.

Class in the Ml module.

Compact fixed-point formatting for live metrics.

A complete OA model file loaded in memory. `ModelFile` owns model configuration, named weights, persistent state, optimizer state, and training progress. `load()` and `save()` implement the native `.oam` wire format. External formats are imported by `oa::ModelTranslator`; they are not alternative representations of this type.

Generic config — universal fields that every architecture has. architecture-specific config follows immediately after (archConfigSize bytes).

Class in the Ml module.

Class in the Ml module.

Class in the Ml module.

Class in the Ml module.

Class in the Ml module.

Model translators are small declarative mapping layers. They inspect the opened source/config assets and emit a complete map; transfer mechanics remain shared and independently tested.

oa::Module — base class for all neural network modules.

oa::ModuleBuffer — persistent/non-trainable buffer registered on a module.

Class in the Ml module.

MultiHeadAttention: Multi-head scaled dot-product self-attention with explicit causal/bidirectional visibility and interchangeable standard/fused causal Flash backends.

GPU momentum + Newton-Schulz5 orthogonalization optimizer. https://github.com/KellerJordan/Muon + https://arxiv.org/abs/2502.16982 Rank-2 parameters use the orthogonalized Muon update. Other ranks use the optimizer's fused GPU momentum update. Muon owns exactly the parameter set supplied by the caller and never delegates to another optimizer.

Class in the Core module.

Sub-modules

Class in the Ui module.

Ready-to-train causal language model: token + position embeddings, a stack of Transformer blocks, final normalization, and vocabulary projection. Input token ids are [batch, contextLength]; logits are [batch*contextLength, vocabSize] for all-position next-token training.

Class in the Ui module.

Smith's 1cycle policy: warmup ramp to maxLr, then cosine anneal to near-zero. initialLr = maxLr/divFactor. finalLr = initialLr/finalDivFactor.

Class in the Core module.

Class in the Core module.

Class in the Core module.

base Optimizer

OPTIMIZER CONFIG — AdamW, SGD, schedulers

No-op optimizer For callers that pass an Optimizer & but update params themselves (e.g. hand-rolled training tutorials using ItTraining purely for cadence + callbacks).

Class in the Core module.

oa::Parameter — named trainable tensor with live gradient access.

Class in the Core module.

Class in the Core module.

Deterministic categorical evaluation over any native OA vector environment. Evaluation is an explicit telemetry boundary: it records the entire horizon first, then performs one execution/synchronization and three compact readbacks.

Complete environment-neutral categorical PPO loop. The environment remains caller-owned; this class owns collection storage, policy bookkeeping, GAE, and optimizer updates. It composes `ItRolloutTraining` because collection and update phases are not the same lifecycle as DQN or SAC.

surface ownership normally stays with the caller. call close() or detachPresentation() before destroying the surface. as misuse containment, destroying an attached presenter transfers the surface and WSI state to the engine; the engine destroys them at close() and the caller must not do so.

Class in the Crypto module.

Class in the Crypto module.

A GPU-resident weight stored in OA's native Q4 or Q8 block encoding. Quantization is deliberately not an `oa::ScalarType`: one logical value owns a packed integer payload and a separate Float32 scale plane. Treating Q4 as a scalar dtype would make matrix byte sizes, strides, views, and element access incorrect. The physical planes remain private and travel as one weight value.

Class in the Core module.

Class in the Ml module.

Relu: ReLU activation layer.

Class in the Ui module.

Class in the Ui module.

generation-safe non-owning handle for one submitted renderer target. Color() remains valid until the exact frame is consumed or abandoned.

Class in the Ui module.

Preallocated circular off-policy storage. Appends and deterministic uniform sampling stay on the GPU; size/cursor are host control metadata and never depend on a tensor readback.

Class in the Ml module.

Class in the Core module.

Class in the Core module.

RmsNorm: RMS normalization layer.

Class in the Ml module.

/ Rnn — vanilla Elman recurrent module. h_new = tanh( W_ih x + b_ih + W_hh h + b_hh ) Mirrors the Gru / GruCell API 1:1 (zeroState / step / forward, stacked layers) so the two recurrent modules are interchangeable in a model. Like the GRU it fuses its pointwise tail (Add + Tanh) into one RnnCellPointwise kernel. The two Linear projections stay separate dispatches; since oa::FnMatrix::Linear is pure dispatch, each needs a manual grad-node attach — see AttachLinearGrad in Rnn.cpp.

Class in the Ml module.

Same-device categorical collector. It owns no environment, network or storage; collect records one complete rollout transaction and returns its exact completion event without waiting.

Rotary position embedding over a flattened [tokens, heads * head_dim] input.

Minimal fixed-alpha SAC trainer. Actor forward returns [B, 2*A] containing mean then log-standard-deviation; each critic consumes [observation, action] concatenated on the last axis and returns [B] or [B,1]. Its independent actor and critic update loops are composed rather than hidden by inheritance.

Class in the Ui module.

Class in the Ui module.

Class in the Ml module.

Class in the Core module.

Class in the Core module.

Class in the Vision module.

Class in the Crypto module.

Class in the Crypto module.

Canonical semantic graph. It is deliberately independent of vulkan and of executable dispatch descriptions so graph compilation can decompose, fuse, partition, place, and schedule without losing public operation semantics.

Deterministic evidence for the provenance relation between semantic operations and executable nodes. One semantic operation with multiple nodes is a decomposition; executable nodes without semantic ownership remain visible compatibility debt while schema migration is incomplete.

Deterministic bridge from a handle-free semantic value to one captured physical-resource identity. resource IDs are assigned by first semantic appearance; vulkan handles and host addresses never enter this descriptor.

Sequential: Sequential container that chains modules together.

Chain N schedulers at milestones. milestones[i] = step at which scheduler i+1 starts. Each sub-scheduler receives step relative to its own start.

SGD

Incremental sponge context — absorb in chunks, squeeze in chunks.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Crypto module.

Silu: SiLU activation layer.

Softmax: Softmax activation layer.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Class in the Core module.

Swiglu: SwiGLU MLP layer with optional weight concatenation.

Class in the Core module.

Class in the Core module.

Manages the device-side coverage atlas and the glyph metadata table.

Class in the Ui module.

Class in the Core module.

Native thread owner. Creation is explicit and fallible; destruction never waits or detaches implicitly. A live thread must be joined or detached by its owning session before this value is destroyed.

Class in the Core module.

Class in the Core module.

BASE TRAINING CONFIG — Shared across LLM, RL, GAN, etc.

Class in the Ml module.

Class in the Ml module.

Class in the Ml module.

Class in the Ui module.

Class in the Ml module.

Class in the Ui module.

Class in the Ui module.

Class in the Core module.

Class in the Core module.

Upsample: Spatial upsampling layer (Nearest or Bilinear).

— global validation controller

Class in the Ml module.

Class in the Core module.

Class in the Core module.

Class in the Ml module.

hardware video decoder session. Wraps VkVideoSessionKHR + DPB management.

Video stream demuxer

hardware video encoder session — wraps VkVideoSessionKHR for encoding.

Video frame resource pool (reuse decoded frames).

Class in the Vision module.

Class in the Vision module.

Class in the Vision module.

Synchronous frame transcoder. The source packet is decoded on the vulkan video queue, converted to RGBA on compute, then submitted to the vulkan encoder without a CPU pixel round-trip. Container demux/mux remains the caller's responsibility.

Windowed or headless application session for inspecting OA images, video, audio and live GPU output. The Viewer owns UI, input, window and presenter state. It either creates one engine for run() or borrows the caller's engine for run(oa::Engine & ); it never creates a second device runtime in that path.

Non-owning live producer attached to the one Viewer application lifecycle. It may render domain-specific overlays, but it never owns the window, swapchain, input pump or presentation submission.

— Passive view description for rendering Declares: camera, target, viewport, and view-specific settings. Does NOT: handle input, own windows, manage swapchains, or animate.

VRAM BUDGET — auto-tune batch_size * seq_len from free VRAM (implementation in budget.cpp)

Class in the Ml module.

Class in the Core module.

Linear warmup to targetLr over warmupSteps, then delegate to after scheduler.

Class in the Core module.

Immutable named-weight source. Implementations may mmap one file, aggregate a sharded manifest, or expose another model container. Returned byte spans remain valid while the source remains alive.

Class in the Core module.

Class in the Core module.

Class in the Core module.