Crate chainrules_core 

Core AD trait definitions (like Julia’s ChainRulesCore.jl).

This crate defines the interface for automatic differentiation without providing an AD engine. It contains:

• Differentiable — tangent space definition for any value type
• ReverseRule — per-operation reverse-mode rule (rrule/pullback)
• ForwardRule — per-operation forward-mode rule (frule/pushforward)
• Error types (AutodiffError, AdResult)
• NodeId, SavePolicy — graph node identifier and save strategy

The AD engine (TrackedTensor, DualTensor, pullback, hvp) lives in the chainrules crate.

Operation-specific AD rules (e.g., einsum rrule/frule) live in the crate that defines the operation.

Examples

Implementing Differentiable for a custom type:

use chainrules_core::Differentiable;

#[derive(Clone)]
struct MyVec(Vec<f64>);

impl Differentiable for MyVec {
    type Tangent = MyVec;
    fn zero_tangent(&self) -> MyVec {
        MyVec(vec![0.0; self.0.len()])
    }
    fn accumulate_tangent(a: MyVec, b: &MyVec) -> MyVec {
        MyVec(a.0.iter().zip(&b.0).map(|(x, y)| x + y).collect())
    }
}

Structs

NodeId

Stable identifier of an AD graph node.

Enums

AutodiffError

AD-specific error type.

SavePolicy

Saved-tensor retention policy for reverse-mode rules.

Traits

Differentiable

Trait defining the tangent space for a differentiable type.

ForwardRule

Forward-mode AD rule interface (frule).

ReverseRule

Reverse-mode AD rule interface (rrule).

Type Aliases

AdResult

Result alias for AD APIs.