Expand description
RONN Core Runtime Engine
This crate provides the foundational components of the RONN (Rust ONNX Neural Network) runtime, including tensor operations, model graph representation, and core execution interfaces.
§Architecture
The core engine follows a layered architecture:
- Types: Fundamental data structures for tensors, graphs, and metadata
- Session: Management of inference sessions and resource isolation
- Tensor: Multi-dimensional array operations with Candle integration
- Graph: Model representation and manipulation utilities
§Example
use ronn_core::{Tensor, DataType, TensorLayout};
// Create a 2x3 tensor with zeros
let tensor = Tensor::zeros(vec![2, 3], DataType::F32, TensorLayout::RowMajor)?;
assert_eq!(tensor.shape(), vec![2, 3]);
assert_eq!(tensor.numel(), 6);Re-exports§
pub use error::CoreError;pub use error::Result;pub use graph::GraphBuilder;pub use graph::GraphStatistics;pub use memory_pool::MemoryPool;pub use memory_pool::PoolConfig;pub use memory_pool::PoolStats;pub use memory_pool::PooledBuffer;pub use memory_pool::global_pool;pub use ops::ArithmeticOps;pub use ops::MatrixOps;pub use ops::ReductionOps;pub use ops::ShapeOps;pub use profiling::CategoryStats;pub use profiling::OperationStats;pub use profiling::ProfileConfig;pub use profiling::ProfileEvent;pub use profiling::ProfileReport;pub use profiling::Profiler;pub use profiling::global_profiler;pub use profiling::init_profiler;pub use session::GlobalStatistics;pub use session::InferenceSession;pub use session::SessionConfig;pub use session::SessionManager;pub use session::SessionStatistics;pub use simd::SimdFeatures;pub use simd::SimdLevel;pub use simd::simd_features;pub use tensor::Tensor;pub use types::AttributeValue;pub use types::CompiledKernel;pub use types::DataType;pub use types::ExecutionProvider;pub use types::GraphEdge;pub use types::GraphNode;pub use types::KernelStats;pub use types::MemoryInfo;pub use types::MemoryType;pub use types::MemoryUsage;pub use types::ModelGraph;pub use types::NodeAttribute;pub use types::NodeId;pub use types::OperatorSpec;pub use types::OptimizationLevel;pub use types::PerformanceProfile;pub use types::ProviderCapability;pub use types::ProviderConfig;pub use types::ProviderId;pub use types::ProviderType;pub use types::ResourceRequirements;pub use types::SessionId;pub use types::SubGraph;pub use types::TensorAllocator;pub use types::TensorBuffer;pub use types::TensorLayout;
Modules§
- error
- Error types for core operations
- graph
- Graph manipulation and validation utilities.
- logging
- Structured logging configuration for RONN.
- memory_
pool - Memory pooling for efficient tensor allocation.
- ops
- Tensor operations module.
- profiling
- Performance profiling infrastructure
- session
- Session lifecycle management for inference contexts.
- simd
- SIMD (Single Instruction Multiple Data) optimizations
- tensor
- Tensor implementation with Candle backend integration.
- types
- Core type definitions for the RONN runtime.
Macros§
- profile
- Profile a scope with the global profiler