Crate ferrum_runtime 

Ferrum Runtime

Device runtime and compute backend implementations for LLM inference.

Re-exports

pub use backends::*;

pub use memory::*;

Modules

backends

Backend implementations for different compute devices

memory

Device memory management implementations

Structs

DefaultBackendRegistry

Default backend registry

Enums

DataType

Data type for tensors

Device

Device type for computation

Traits

ComputeBackend

Compute backend for tensor operations and kernel execution

DeviceMemoryManager

Device memory manager for raw memory operations

KernelExecutor

Kernel executor for custom GPU kernels

TensorFactory

Tensor factory for creating tensors on specific backends

TensorLike

Core tensor trait for zero-copy, device-aware operations

TensorOps

Basic tensor operations

Functions

global_backend_registry

Get the global backend registry

Type Aliases

Result

Result type used throughout Ferrum

TensorRef

Reference-counted tensor handle for zero-copy sharing