vllm.model_executor.kernels.attention.dsa.dcp_indexer_cutedsl
¶
Functions:
-
dcp_topk_id_bits–Bits that hold every candidate token id (a position below max_model_len).
-
stable_topk_from_gathered_candidates_cutedsl–score_bits: high score bits that rank (16 for upcast BF16 logits).
_dcp_merge_topk_and_interleave(vllm_config)
¶
(topk, cp_interleave) the DCP merge runs with.
Kpool indexers (GLM-5.x) merge pool ids: topk and the interleave are both
in units of index_kpool tokens.
Source code in vllm/model_executor/kernels/attention/dsa/dcp_indexer_cutedsl.py
dcp_topk_id_bits(max_model_len)
¶
Bits that hold every candidate token id (a position below max_model_len).
stable_topk_from_gathered_candidates_cutedsl(gathered, topk, out=None, score_bits=32, id_bits=32)
¶
score_bits: high score bits that rank (16 for upcast BF16 logits).
id_bits: bits that hold every token id.