flashinfer.gdn_paged.gdn_paged_prefill_slots

flashinfer.gdn_paged.gdn_paged_prefill_slots(cu_seqlens: Tensor, page_indices: Tensor, page_indptr: Tensor, page_size: int, last_page_len: Tensor) → GDNPagedPrefillSlots

Pool rows that chunk_gated_delta_rule_paged() derives from a page table, computed in one launch without a host synchronization.

The rows depend only on the batch (cu_seqlens and the page table), not on the layer, so a model can compute them once per step and pass them to every layer’s call as slots.

Parameters:

• cu_seqlens – As in chunk_gated_delta_rule_paged().

• page_indices – As in chunk_gated_delta_rule_paged().

• page_indptr – As in chunk_gated_delta_rule_paged().

• page_size – As in chunk_gated_delta_rule_paged().

• last_page_len – As in chunk_gated_delta_rule_paged().

Returns:

[num_seqs] int32 read rows, final-state rows, first checkpoint entries and prefix lengths.

Return type:

GDNPagedPrefillSlots