Question about split_kv_cache in paged_attn.py

[March-H]

@staticmethod
    def split_kv_cache(
        kv_cache: torch.Tensor,
        num_kv_heads: int,
        head_size: int,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        x = 16 // kv_cache.element_size()
        num_blocks = kv_cache.shape[1]

        key_cache = kv_cache[0]
        key_cache = key_cache.view(num_blocks, num_kv_heads, head_size // x,
                                   -1, x)
        print(key_cache.shape)
        value_cache = kv_cache[1]
        value_cache = value_cache.view(num_blocks, num_kv_heads, head_size, -1)
        return key_cache, value_cache

Why does the key_cache need to be split like this, but the value_cache doesn't? What is the role of x?
I’m curious about this way of splitting, but I can’t figure out the answer. Is this related to the storage format in paged_attn? I hope someone can help clarify my confusion. Thank you very much!:)