[CI] CPU kernel benchmark for ngram_match — DO NOT MERGE

[cloudforge1]

Motivation

PR #7136 benchmarks the GPU ngram_match kernel but the "CPU path" column
only measures D2H/H2D tensor copy overhead, not the actual C++ kernel
computation. This makes the reported speedup (14×–1,700×) misleading — the
real GPU-vs-CPU-compute speedup is much more modest (~0.3×–5.8× per NKNaN's
profiling data).

This PR adds a standalone CPU benchmark that calls the production C++
kernel (ngram_match.cc / find_candidate_pred_tokens) with CPU-placed
tensors, using the same 5-group experiment dimensions so the numbers are
directly comparable.

⚠️ NOT FOR MERGE — this is a reference-data-only PR. The .cc file
is deleted in the GPU kernel branch; this benchmark exists on develop
where both .cc and .cu coexist.

Modifications

• Added tests/spec_decode/test_benchmark_ngram_cpu.py (354 lines)
  • 5 groups matching GPU benchmark dimensions (seq_len, batch_size, hit type, threshold, threshold×batch)
  • 2 latency tests (standard + extreme) matching GPU benchmark configs
  • Adaptive run counts (100–1000) to stay within 3-minute total runtime
  • Uses paddle.CPUPlace() tensors → dispatches to .cc C++ kernel

Usage or Command

cd FastDeploy
python tests/spec_decode/test_benchmark_ngram_cpu.py

Accuracy Tests

Not applicable — benchmark-only, no functional changes.

Checklist

• pre-commit hooks pass (black, isort, flake8, ruff)
• Same API signature as GPU benchmark for 1:1 comparison
• Adaptive run counts to avoid CI timeout (est. ~2.2 min total)
• NOT FOR MERGE — reference data only

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@da3dfe1). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #7203   +/-   ##
==========================================
  Coverage           ?   74.19%           
==========================================
  Files              ?      376           
  Lines              ?    52941           
  Branches           ?     8260           
==========================================
  Hits               ?    39279           
  Misses             ?    10910           
  Partials           ?     2752           

Flag	Coverage Δ
GPU	74.19% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.