pub fn sigmoid_blend_device(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
output: DevicePtr,
src: DevicePtr,
gate_ptr: DevicePtr,
num_elements: u32,
stream: u64,
) -> Result<()>Expand description
2026-09-25: output[i] += sigmoid(*gate_ptr) * src[i], where gate_ptr is
one BF16 scalar in device memory (bf16_sigmoid_blend_device in
kernels/gb10/common/residual_add.cu).