[None][docs] fix legacy benchmark KV cache config

[imitater-dou]

Summary

• nest the legacy benchmark's forced KV cache dtype under kv_cache_config
• name the accepted kv_cache_config.dtype field in the adjacent tip

Why

tensorrt_llm/bench/benchmark/utils/general.py reads the YAML through llm_args_dict.get(\"kv_cache_config\", {\"dtype\": \"auto\"}) and then reads dtype from that mapping. The documented top-level kv_cache_dtype is therefore ignored and the benchmark falls back to auto.

Test plan

• Reproduced the current parser expression locally: {\"kv_cache_dtype\": \"fp8\"} resolves to auto, while {\"kv_cache_config\": {\"dtype\": \"fp8\"}} resolves to fp8
• git diff --check

Signed-off-by: ADou ikun3.1415927@gmail.com

Dev Engineer Review

• Updated the PyTorch workflow example to nest the cache dtype under kv_cache_config.dtype.
• Documented the accepted values: auto, fp8, and nvfp4.
• The change matches the parser configuration path and prevents an unintended fallback to auto.
• No public entities, test files, or configuration files changed.

QA Engineer Review

No test changes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 9bd4bb74-6cbf-414e-9447-0f5050d7c420

📥 Commits

Reviewing files that changed from the base of the PR and between 3901bca and c729c5e.

📒 Files selected for processing (1)

• docs/source/legacy/performance/perf-benchmarking.md

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

Walkthrough

The benchmarking documentation replaces the legacy kv_cache_dtype setting with kv_cache_config.dtype. The documented values remain auto, fp8, and nvfp4.

Changes

Quantization documentation

Layer / File(s)	Summary
Update KV cache dtype example docs/source/legacy/performance/perf-benchmarking.md	The PyTorch workflow example uses nested kv_cache_config.dtype configuration and retains the valid values auto, fp8, and nvfp4.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Merge Risk: ⚪ Minimal · up to c729c

The benchmark documentation now shows the KV-cache dtype in the configuration structure consumed by the runtime, preventing the example from silently defaulting to auto. The change is ready to merge.

Suggested reviewers: bowenfu

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies a documentation fix for the legacy benchmark KV cache configuration. It is concise and related to the main change.
Description check	✅ Passed	The description explains the issue, the parser behavior, the documentation change, and the validation performed. It does not include the PR checklist section, but the core required information is comp…
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

Full details: Description check

Explanation

The description explains the issue, the parser behavior, the documentation change, and the validation performed. It does not include the PR checklist section, but the core required information is complete.

Full details: Docstring Coverage

Explanation

No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0 files. (1 skipped: 1 unsupported.)

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}