Struct OptimizerState 

pub struct OptimizerState {
    pub optimizer_type: String,
    pub scalars: HashMap<String, f64>,
    pub buffers: HashMap<String, Vec<f64>>,
}

A serializable snapshot of an optimizer’s internal state.

Contains named scalar values (e.g., step count, learning rate) and named f64 buffers (e.g., momentum vectors, second moment estimates).

This follows the PyTorch state_dict() / load_state_dict() pattern but is Rust-native and format-agnostic.

Fields

optimizer_type: String

Optimizer type name (e.g., “Adam”, “SGD”) for validation on load.

scalars: HashMap<String, f64>

Named scalar values (step count, hyperparameters, etc.)

buffers: HashMap<String, Vec<f64>>

Named f64 buffers (momentum vectors, second moment estimates, etc.) Each buffer is flattened to a single Vec; the key encodes the parameter index: e.g., “m.0”, “m.1”, “v.0”, “v.1”.

Implementations

impl OptimizerState

pub fn new(optimizer_type: impl Into<String>) -> OptimizerState

Create an empty state dict for the given optimizer type.

pub fn set_scalar(&mut self, key: impl Into<String>, value: f64)

Insert a scalar value.

pub fn set_buffer(&mut self, key: impl Into<String>, data: Vec<f64>)

Insert a buffer.

pub fn get_scalar(&self, key: &str) -> Option<f64>

Get a scalar value.

pub fn get_buffer(&self, key: &str) -> Option<&Vec<f64>>

Get a buffer.

Trait Implementations

impl Clone for OptimizerState

fn clone(&self) -> OptimizerState

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for OptimizerState

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.