Struct AttnV41LayerState 

pub struct AttnV41LayerState { /* private fields */ }

2026-09-25: One sequence’s attention state for one layer, across prefill and decode: the bf16 window ring [window, hd]; on kv sources also the compressor’s partial group (f32 kv and score, [ratio, hd] each, used at ratio > 1), the bf16 latent cache [max_seq / ratio, hd] and the bf16 index-key cache [max_seq / ratio, index_hd].

Implementations

impl AttnV41LayerState

pub fn new( gpu: &dyn GpuBackend, c: &AttnV41Cfg, role: LayerRole, ) -> Result<Self>

pub fn window(&self) -> DevicePtr

2026-09-25: The window ring, bf16 [window, hd]: a pointer the captured decode steps bake.

pub fn free(&mut self, gpu: &dyn GpuBackend) -> Result<()>

2026-09-25: Free this sequence’s device buffers (the window ring and, on kv sources, the two compressor-state buffers and both caches) and clear the pointers. A second call frees nothing: the window is NULL (GpuBackend::free ignores NULL) and the options are None.