Crate realizar 

realizar has moved to aprender-serve.

This crate re-exports aprender-serve for backward compatibility. New code should depend on aprender-serve directly.

Modules

api

HTTP API for model inference

apr

Aprender .apr format support (PRIMARY inference format)

apr_transformer

APR Transformer format for WASM-compatible LLM inference

arch_requirements

Per-architecture required weight roles (GH-279).

audit

Audit trail and provenance logging

bench

Benchmark harness for model runner comparison

bench_preflight

Preflight validation protocol for deterministic benchmarking

bench_viz

Benchmark visualization for inference comparison (PAR-040)

brick

ComputeBrick architecture for token-centric, self-verifying inference

cache

Model caching and warming for reduced latency

capability

GH-280: Kernel capability gate — contract-driven GPU admission control.

chat_template

Chat template engine for model-specific message formatting

cli

CLI command implementations (extracted for testability) CLI command implementations

contract_gate

GGUF to APR Transformer converter

convert

GGUF to APR Transformer Converter

cuda

CUDA PTX generation for NVIDIA GPUs

error

Error types for Realizar

explain

Model explainability (SHAP, Attention)

format

Unified model format detection (APR, GGUF, SafeTensors)

generate

Text generation and sampling strategies

gguf

GGUF (GPT-Generated Unified Format) parser

gpu

GPU acceleration module (Phase 4: ≥100 tok/s target)

grammar

Grammar-constrained generation for structured output

infer

High-level inference API for CLI tools

inference

SIMD-accelerated inference engine using trueno

inference_trace

Inference tracing for debugging LLM pipelines

layers

Neural network layers for transformer models

memory

Memory management for hot expert pinning

metrics

Metrics collection and reporting for production monitoring

model_loader

Unified model loader for APR, GGUF, and SafeTensors

moe

Mixture-of-Experts (MOE) routing with Capacity Factor load balancing

observability

Observability: metrics, tracing, and A/B testing

paged_kv

PagedAttention KV cache management

parallel

Multi-GPU and Distributed Inference

ptx_parity

PTX Parity Validation — GH-219

quantize

Quantization and dequantization for model weights

registry

Model registry for multi-model serving

safetensors

Safetensors parser

safetensors_cuda

SafeTensors CUDA inference (PMAT-116)

safetensors_infer

SafeTensors inference support (PAR-301)

scheduler

Continuous batching scheduler

serve

Aprender ML Model Serving API

speculative

Speculative decoding for LLM inference acceleration

stats

Statistical analysis for A/B testing with log-normal latency support

target

Multi-target deployment support (Lambda, Docker, WASM) Multi-Target Deployment Support

tensor

Tensor implementation

tensor_names

GH-311: Contract-driven tensor name resolution (tensor-names-v1.yaml codegen)

tokenizer

Tokenizer for text encoding and decoding

tui

TUI monitoring for inference performance TUI Monitoring for LLM Inference

uri

Pacha URI scheme support for model loading Pacha URI scheme support for model loading

viz

Benchmark visualization using trueno-viz.

warmup

Model warm-up and pre-loading Model Warm-up and Pre-loading

Macros

profile_clear

Clear the thread-local profiler

profile_report

Get a report from the thread-local profiler

profile_start

Start profiling an operation using the thread-local profiler

profile_stop

Stop profiling an operation using the thread-local profiler

trace_cpu

No-op version of trace_cpu when “trace” feature is disabled.

trace_gpu

No-op version of trace_gpu when “trace” feature is disabled.

Structs

BatchInferenceConfig

Batch inference: load model once, process multiple prompts sequentially.

BatchPrompt

Single prompt in a batch

BatchResult

Single result from batch inference

BatchStats

Aggregate stats for the entire batch

InferenceConfig

Configuration for inference

InferenceResult

Result from inference

InferenceTracer

Inference tracer

KernelDimensions

Model dimensions needed to construct and validate kernels

MappedSafeTensorsModel

Zero-copy memory-mapped SafeTensors model container

ModelInfo

Model information for trace header

PreparedTokens

Tokenized input that has been processed through chat template formatting.

PtxParityReport

Full PTX parity validation report

SafetensorsConfig

Model configuration from config.json

ShardedSafeTensorsModel

Sharded SafeTensors model container (GH-213)

Tensor

N-dimensional tensor with automatic backend dispatch

TraceConfig

Trace configuration

ValidatedAprTransformer

Validated APR Transformer — compile-time guarantee of tensor data quality

Enums

RealizarError

Error type for all Realizar operations

TraceStep

Inference pipeline steps (State Machine states per AWS Step Functions model)

WeightRole

Weight roles that may be required for a transformer layer. Each architecture requires a subset of these.

Constants

VERSION

Library version

Functions

required_roles

Returns the required weight roles for a given architecture.

rms_norm

RMSNorm (Root Mean Square Layer Normalization)

rms_norm_into

RMSNorm to pre-allocated buffer (zero-allocation path)

run_batch_inference

Run batch inference, auto-detecting model format (GGUF or APR).

run_inference

Run inference on a model

Type Aliases

Result

Result type alias for Realizar operations