pub fn l2_norm(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
data: DevicePtr,
num_heads: u32,
head_dim: u32,
eps: f32,
num_tokens: u32,
stride: u32,
stream: u64,
) -> Result<()>Expand description
2026-09-25: In-place L2 normalization of each head, x / sqrt(sum(x^2) + eps) (kernel
l2_norm_bf16). data is [num_tokens, stride] BF16 with num_heads * head_dim
used per token; one block per (head, token).