Method process_kv

cake-core/src/models/common/cache.rs:163–170

Process the input k and v by either generating their cache entry or applying a previously cached one.

(
        &mut self,
        block_idx: usize,
        k: Tensor,
        v: Tensor,
    )

    /// Process the input k and v by either generating their cache entry or applying a previously cached one.
    pub fn process_kv(
        &mut self,
        block_idx: usize,
        k: Tensor,
        v: Tensor,
    ) -> Result<(Tensor, Tensor)> {
        self.process_kv_inner(block_idx, k, v, self.max_seq_len)
    }

    /// Like `process_kv` but caps the KV cache to `window` tokens (sliding window attention).
    pub fn process_kv_windowed(