Method matmul

cake-core/src/backends/mod.rs:185–187

Matrix multiplication. GPU backends override to use accelerated matmul. Default: delegates to candle's CPU/CUDA/Metal matmul.

(&self, a: &Tensor, b: &Tensor)

    /// Matrix multiplication. GPU backends override to use accelerated matmul.
    /// Default: delegates to candle's CPU/CUDA/Metal matmul.
    fn matmul(&self, a: &Tensor, b: &Tensor) -> Result<Tensor> {
        a.matmul(b)
    }

    /// Pre-process a linear weight for optimal matmul performance.
    /// Called once at model load time. Backends can pre-convert dtype,