Skip to main content

preload_resident_blob

Function preload_resident_blob 

Source
pub fn preload_resident_blob(key: u64, bytes: &[u8]) -> bool
Expand description

Upload an immutable prepared-plan blob to the permanent CUDA slab.

The caller supplies the stable content key. This is used for small f32 vectors such as RMSNorm weights as well as quantized matrices, so the token path never performs H2D writes.