[TRTLLM-12112][feat] Support v2 KV cache stats

[yizhang-nv]

@coderabbitai summary

Description

Port V2 KV cache stats support onto the DeepSeek V4 feature branch.

This PR adds V2 KV cache stats delta tracking and wires it into the existing PyExecutor stats path so LLM.get_stats() and server polling can surface kvCacheStats / kvCacheIterationStats for KV cache manager V2. It also adds V1/V2 alignment coverage for basic block counts, SWA windows, and stacked multi-window reuse scenarios.

Test Coverage

• python3 -m py_compile tests/integration/defs/kv_cache/test_kv_cache_iteration_stats_alignment.py
• git diff --check -- tests/integration/defs/kv_cache/test_kv_cache_iteration_stats_alignment.py
• /home/yizhan/.local/bin/pre-commit run --files tests/integration/defs/kv_cache/test_kv_cache_iteration_stats_alignment.py
• B200 LLM_MODELS_ROOT=/scratch.trt_llm_data/llm-models pytest -q -s tests/integration/defs/kv_cache/test_kv_cache_iteration_stats_alignment.py
  • 3 passed, 3 warnings in 153.30s
  • log: /home/scratch.yizhan_sw_1/logs/2026-05-08/6u1g-0015/LLM_MODELS_ROOT__scratch_trt_llm_data_llm-models___01-35-50.stdout.log

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

[tensorrt-cicd]

PR_Github #49147 [ run ] triggered by Bot. Commit: bd74724 Link to invocation

[tensorrt-cicd]

PR_Github #49133 [ run ] completed with state ABORTED. Commit: 107553d

Link to invocation

[tensorrt-cicd]

PR_Github #49147 [ run ] completed with state SUCCESS. Commit: bd74724
/LLM/main/L0_MergeRequest_PR pipeline #38832 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[nvpohanh]

@lowsfer could you review this? thanks

[yizhang-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #49308 [ run ] triggered by Bot. Commit: 9ec657f Link to invocation

[tensorrt-cicd]

PR_Github #49308 [ run ] completed with state SUCCESS. Commit: 9ec657f
/LLM/main/L0_MergeRequest_PR pipeline #38970 completed with status: 'SUCCESS'

CI Report

Link to invocation