feat: Support stateful responses API with previous_response_id chaining (cherry-pick to release/0.13.0)

[ajcasagrande]

Summary

• Cherry-pick of ea33cc5a57 from main into release/0.13.0
• Original PR: feat: Support stateful responses API with previous_response_id chaining #1347 — feat: Support stateful responses API with previous_response_id chaining

Conflicts

None — clean cherry-pick onto prerequisite base.

Stack

This is PR 5 of 6. Merge after #1377, #1378, #1371, and #1372.

1. fix(mmap): make get_conversation thread-safe, prefault pages, drop executor hop (#1245) #1377 — fix(mmap): make get_conversation thread-safe, prefault pages, drop executor hop #1245 (mmap thread-safety)
2. fix(dataset): restore random_pool batch-size support with --input-file (#1274) #1378 — fix(dataset): restore random_pool batch-size support with --input-file #1274 (random_pool batch-size)
3. fix(config): stop silently dropping CLI flags passed with --config (cherry-pick to release/0.13.0) #1371 — fix(config): stop silently dropping CLI flags passed with --config #1280 (config flag routing)
4. fix(test-harness): implement num_prompt_special_tokens on FakeTokenizer (cherry-pick to release/0.13.0) #1372 — fix(test-harness): implement num_prompt_special_tokens on FakeTokenizer #1322 (FakeTokenizer)
5. This PR — feat: Support stateful responses API with previous_response_id chaining #1347 (stateful responses / previous_response_id)
6. feat(kubernetes): add native Kubernetes execution + prerequisite stack (cherry-pick #1245+#1274+#1280+#1322+#1325+#1326+#1347 to release/0.13.0) #1379 — combined feat: control-plane groundwork for k8s execution — dead-component detection, result finalization, ZMQ reconnect #1325+ feat(k8s): AIPerf goes k8s-native — operator, CRDs, JobSet execution, and the aiperf kube CLI #1326 (k8s-native)

[github-actions]

Try out this PR

Quick install:

pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@6a3c581f4496346ce3ce85ee9dcff5c0afdd157b

Recommended with virtual environment (using uv):

uv venv --python 3.12 && source .venv/bin/activate
uv pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@6a3c581f4496346ce3ce85ee9dcff5c0afdd157b

---

Last updated for commit: 6a3c581 • Browse code

[github-actions]

Fern Docs Preview: https://nvidia-preview-01a05db0-8f1c-70dd-b1ed-c6355ac23a46.docs.buildwithfern.com/aiperf/dev

[codecov]

Codecov Report

❌ Patch coverage is 93.22034% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/aiperf/endpoints/openai_responses.py	89.18%	2 Missing and 2 partials ⚠️

📢 Thoughts on this report? Let us know!