Crate openinfer_simulator 

OpenInfer

OpenInfer is an inference graph and execution framework for machine-learning workloads. It provides a graph model, a runtime, and tensor utilities that can execute on CPU (and optionally Vulkan) backends.

Quick start

use openinfer::{graph, Device, ModelLoader, Simulator};
let model = ModelLoader::default();
let g = graph! {
    block main {
        // ... graph nodes ...
    }
};
let sim = Simulator::new(&model, &g, Device::Cpu)?;
let _executor = sim.with_trace().make_executor()?;

Key concepts

• Graph and Node describe the computation structure.
• Tensor and TensorValue hold data with explicit shapes and dtypes.
• Simulator and Executor manage validation and execution.
• ModelLoader provides external state and parameters.

Module map

• graph: graph nodes, blocks, and serialization.
• runtime: executor, validation, and tracing.
• tensor: tensor containers and dtype utilities.
• ops: op registry and kernel dispatch.

Modules

logging

Macros

critical

Emit a critical message via the logging subsystem.

error

Emit an error message via the logging subsystem.

fetch_executor

graph

Build an OpenInfer Graph from the DSL input.

insert_executor

log

Emit a raw log message using println!.

trace

Emit a trace message via the logging subsystem.

try_fetch_executor

try_insert_executor

vk_trace

Emit a Vulkan trace message via the logging subsystem.

warning

Emit a warning message via the logging subsystem.

Structs

BF16

Brain-float16 (BF16) scalar storage.

Bitset

Packed bitset storage type.

Block

A named block of nodes (control-flow unit).

CacheAccess

Describes a cache access pattern in a node.

Executor

Executes a graph against a model and mutable runtime state.

F8

8-bit float storage (custom format).

F16

IEEE 754 half-precision (F16) scalar storage.

Graph

Graph structure containing blocks and variable declarations.

GraphDeserialize

Deserialize graphs from JSON representations.

GraphSerialize

Serialize graphs into JSON representations.

I1

1-bit signed integer storage.

I2

2-bit signed integer storage.

I4

4-bit signed integer storage.

ModelLoader

Loads .oinf model files and exposes tensors/metadata.

Node

A graph node with index and kind.

OpAttr

Named attribute for an op invocation.

OpAttrs

Collection of op attributes.

Random

Random tensor generator for a specific element type.

Simulator

High-level entry point for validating and executing a graph.

T1

1-bit ternary storage.

T2

2-bit ternary storage.

Tensor

Owned tensor container with shape and stride metadata.

TensorOptions

Tensor construction options (shape/stride overrides).

Timer

Simple per-thread timing utility for runtime instrumentation.

TraceEvent

Execution trace record for a single node.

U1

1-bit unsigned integer storage.

U2

2-bit unsigned integer storage.

U4

4-bit unsigned integer storage.

VarDecl

Variable declaration within a graph.

VarInfo

Metadata about a variable stored in a model file.

Enums

AttrValue

Attribute value used by ops in the graph.

CacheIndexExpr

Cache index expression (single or slice).

CacheIndexValue

Index literal or identifier for cache addressing.

DType

Supported element dtypes.

Device

Execution device selection for a simulation run.

MemoryKind

Variable storage classification.

NodeKind

Node variants that make up a graph.

OpKind

Operation kind supported by the runtime.

ScalarValue

Scalar value that can be expanded into tensors.

TensorValue

Runtime tensor value with an enum over concrete dtypes.

TraceEventKind

Kind of trace event emitted during execution.

TypeRule

Type inference rule for op outputs.

Traits

Fetchable

Adapter for fetching runtime values by name.

FormatValue

Format a value for compact human-readable output.

TensorElement

Element type that can be converted to/from TensorValue.

Functions

format_truncated

Format a slice with head/tail truncation.

op_schema

Lookup the schema for a specific op kind.