pub fn moe_batched_blend(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
output: DevicePtr,
shared_out: DevicePtr,
normed: DevicePtr,
gate_weight: DevicePtr,
hidden_size: u32,
num_tokens: u32,
stream: u64,
) -> Result<()>Expand description
Batched sigmoid blend: output += sigmoid(dot(normed, gate_weight)) * shared_out.
Grid: (num_tokens, 1, 1) Block: (256, 1, 1)