Struct TrainSpec 

pub struct TrainSpec {
    pub graph: Graph,
    pub params: Vec<WeightRef>,
    pub grad_start: usize,
    pub loss_index: usize,
    pub data: Vec<DataRef>,
    pub seed_input: Option<String>,
    pub momentum: f32,
    pub lr_per_epoch: Vec<f32>,
    pub batch: usize,
    pub device: String,
    pub grad_group: u64,
    pub push_data: bool,
}

A self-describing training job for a generic worker: everything needed to train a model the worker has no code for.

Fields

graph: Graph

Backward graph. Outputs are [.. , grad_0, grad_1, …], with the gradients starting at grad_start and matching params 1:1.

params: Vec<WeightRef>

Trainable params + where to fetch their initial values (node-local URI).

grad_start: usize

Index in graph.outputs where the gradients begin (params[i] ↔ outputs[grad_start + i]).

loss_index: usize

Index in graph.outputs of the scalar loss (for logging).

data: Vec<DataRef>

Inputs fed each step from node-local data shards.

seed_input: Option<String>

Name of the backward seed input to feed 1.0 (e.g. "d_output"), if any.

momentum: f32

SGD momentum.

lr_per_epoch: Vec<f32>

Learning rate per epoch (len() = number of epochs). The master owns the schedule; the worker just applies lr_per_epoch[epoch].

batch: usize

Mini-batch size (matches the graph’s input batch dim).

device: String

Device directive: auto (node’s fastest), or a name (metal, cuda, …).

grad_group: u64

The collective group id the worker’s reduce closure averages over (informational — the closure owns the actual all-reduce).

push_data: bool

If set, the master pushes each worker its data shard over the wire (for nodes with no shared filesystem); the worker rewrites its DataRefs to the received local files. When false, workers resolve the URIs node-locally (shared FS / pre-staged data).

Trait Implementations

impl Clone for TrainSpec

fn clone(&self) -> TrainSpec

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TrainSpec

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for TrainSpec

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for TrainSpec

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.