Add SPEED-Bench reference synthetic AL values for DeepSeek-V4-Pro MTP 1-8

[qiching]

Measured with SPEED-Bench coding dataset, temperature=1.0, thinking=true. Values used for synthetic acceptance rate configuration in MTP benchmarks.

AL Values

MTP	AL
1	1.90
2	2.60
3	2.97
4	3.04
5	3.13
6	3.08
7	3.13
8	3.12

---

Note

Low Risk
Documentation-style benchmark reference data only; no runtime or production code paths change.

Overview
Adds benchmarks/speedbench-reference-al.yaml with measured acceptance length (AL) targets for DeepSeek-V4-Pro at MTP levels 1–8, keyed by num_speculative_tokens.

Values come from SPEED-Bench (coding dataset, temperature 1.0, thinking on) and are intended as golden AL inputs for synthetic acceptance rate setup in MTP benchmarks.

^{Reviewed by Cursor Bugbot for commit 1de285d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[xinli-sw]

@qiching , a few recommendations

1. title [1/N] Synthetic MTP - Add SPEED-Bench reference synthetic AL values for DeepSeek-V4-Pro MTP 1-8

2. In the PR description, mention that we will also have speedbench as part of github workflows, however, as the first iteration, we'd like to get alignment to make sure partners all feel confident and equally about these AL values

3. attach full repro for the numbers (serve command, installation, speedbench command, etc)

4. attach full results of the runs you had (the jsonl file) for audibility

Great work so far, cc @benchislett @functionstackx