flashinfer.page#

Kernels to manipulte paged kv-cache.

Append new K/V tensors to Paged KV-Cache#

append_paged_kv_cache(append_key, ...[, ...])

Append a batch of key-value pairs to a paged key-value cache.

get_batch_indices_positions(append_indptr, ...)

Convert append indptr and sequence lengths to batch indices and positions.