flashinfer.gdn_paged.GDNPagedPrefillSlots¶

class flashinfer.gdn_paged.GDNPagedPrefillSlots(state_indices: Tensor, output_state_indices: Tensor, checkpoint_cu_starts: Tensor, prefix_lens: Tensor)¶

Per-sequence pool rows of a paged prefill batch, all [num_seqs] int32. See gdn_paged_prefill_slots().

__init__()¶

Methods

`__init__`()
`count`(value, /)	Return number of occurrences of value.
`index`(value[, start, stop])	Return first index of value.

Attributes

`checkpoint_cu_starts`	`page_indices` entry of the first page boundary the new tokens cross.
`output_state_indices`	Last page, written with the final state.
`prefix_lens`	Tokens before the new ones.
`state_indices`	Page holding the last prefix token, read as the initial state (`-1`: empty prefix, zero state).