flashinfer.get_dcp_spec_counter_bytes

flashinfer.get_dcp_spec_counter_bytes(batch_size: int, q_len_per_req: int, num_kv_heads: int) int

Bytes for the v4 completion tickets, zeroed once then self-reset.

Parameters:
  • batch_size (int) – Number of requests in the batch.

  • q_len_per_req (int) – Number of speculative query rows per request.

  • num_kv_heads (int) – Number of rank-local KV heads.

Returns:

Required caller-owned completion-buffer size in bytes.

Return type:

int