pub fn kv_len_ptr_table(
e: &Engine,
cache: &Cache,
pos_ctr: Option<&CudaSlice<i32>>,
) -> Result<CudaSlice<u64>, Box<dyn Error>>Expand description
Per-layer kvl.len_d device-pointer table (+ the position counter appended when
pos_ctr is given) for spec_rollback_stream. Pointers are stable for the cache’s
lifetime; 0 marks layers without KV (linear-attention / KV-shared).