For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Python function

spec_decode_cache_slack

spec_decode_cache_slack()

max.nn.kv_cache.spec_decode_cache_slack(params)

source

Computes the extra KV positions a request may occupy past max_seq_len.

A speculative-decode step can over-speculate past the per-request max_seq_len cap into this slack (the KV pool reserves it beyond max_seq_len), so any per-request sizing derived from max_seq_len – the pool’s page budget, the sparse-indexer score scratch, etc. – must add it. Centralized here so _compute_seq_len, pool sizing, and OverlapTextGenerationPipeline._effective_max_cache_length stay in sync.

Parameters:

params (KVCacheParamInterface) – The KV cache parameters. The speculative-decoding fields (num_draft_tokens and num_draft_tokens_per_step) determine the slack.

Returns:

The number of extra KV positions to reserve past max_seq_len, or 0 when speculative decoding is off.

Return type:

int

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit