Struct ColumnParallelLinear 

pub struct ColumnParallelLinear { /* private fields */ }

Splits the output dimension across n_workers.

With n_workers = W and output size N, each worker holds weights of shape [n_in, N/W] and bias of shape [N/W].

Implementations

impl ColumnParallelLinear

pub fn new( n_in: usize, n_out: usize, config: TensorParallelConfig, seed: u64, ) -> NeuralResult<Self>

Create a column-parallel linear layer.

Errors

• n_out is not divisible by config.n_workers.
• config.n_workers == 0.

pub fn forward(&self, input: &Array2<f64>) -> NeuralResult<Array2<f64>>

Forward pass.

Each worker computes y_i = input @ W_i + b_i. If gather_output, the results are concatenated to [batch, n_out]; otherwise only the first worker’s output is returned (for single-process simulation with gather_output = false).

pub fn n_out(&self) -> usize

Total output features (after all-gather).

pub fn n_workers(&self) -> usize

Number of simulated workers.