flashinfer.gdn_paged.gdn_paged_decode_slots¶
- flashinfer.gdn_paged.gdn_paged_decode_slots(page_indices: Tensor, page_indptr: Tensor, last_page_len: Tensor) GDNPagedDecodeSlots¶
Pool rows that
gated_delta_rule_decode_paged()derives from a page table, computed in one launch without a host synchronization. Compute them once per decode step and pass them to every layer’s call asslots.- Parameters:
page_indices – As in
gated_delta_rule_decode_paged().page_indptr – As in
gated_delta_rule_decode_paged().last_page_len – As in
gated_delta_rule_decode_paged().
- Returns:
[B]int32 read rows (the previous page when the decoded token opens a new page, else the last page) and write rows (the last page).- Return type: