Add autotuner for benchmark parameter optimization

[Copilot]

The benchmark framework measures QPS/latency/recall for given configurations but lacks automated parameter tuning. This adds a tool to sweep parameters and identify optimal configurations.

Implementation

New binary: diskann-tools/bin/autotuner

• Parameter sweep engine for max_degree (R), l_build, search_l, num_pq_chunks
• Generates all build parameter combinations, runs benchmarks, parses results
• Three optimization criteria: maximize QPS at target recall, minimize latency at target recall, maximize recall
• Outputs individual configs/results plus summary with best configuration

CLI commands:

# Generate example sweep config
autotuner example --output sweep_config.json

# Run sweep
autotuner sweep \
  --base-config benchmark.json \
  --sweep-config sweep_config.json \
  --output-dir ./results \
  --criterion qps \
  --target-recall 0.95

Configuration format:

{
  "max_degree": [16, 32, 64],
  "l_build": [50, 75, 100],
  "search_l": [10, 20, 30, 40, 50],
  "num_pq_chunks": [8, 16, 32]
}

Changes

• Add diskann-tools/src/bin/autotuner.rs - main implementation with tests
• Add diskann-tools/AUTOTUNER.md - usage documentation
• Update diskann-benchmark/README.md - reference autotuner
• Update diskann-tools/Cargo.toml - add serde_json dependency

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Create an autotuner tool on top of benchmark</issue_title>
<issue_description>## Is your feature request related to a problem? Please describe.
the repo contains a comprehensive benchmark framework that takes configuration as inputs and measures throughput (qps), latency etc. We want a tool that builds on this to sweep over a subset of parameters, for example R (graph degree), l_build, l_search, and quantization bytes where applicable, and identify the best configuration.

Describe the solution you'd like

A clear and concise description of what you want to happen.

Describe alternatives you've considered

A clear and concise description of any alternative solutions or features you've considered.

Provide references (if applicable)

If your feature request is related to a published algorithm/idea, please provide links to
any relevant articles or webpages.

Additional context

Add any other context or screenshots about the feature request here.

</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Create an autotuner tool on top of benchmark #735

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.