[None][feat] add DSV4 KV cache pool ratio config

[jiaganc]

@coderabbitai summary

Description

Add DeepSeek-V4 KV cache configuration knobs for KV cache manager v2:

• kv_cache_config.pool_ratio: user-provided initial pool ratios. When set for DSV4, this is passed to KV cache manager v2 as initial_pool_ratio and bypasses typical-step and constraint-based initial sizing.
• kv_cache_config.avg_seq_len: average sequence length used by DSV4 typical-step construction. When unset, DSV4 continues to use max_seq_len.

The change also validates pool ratios, keeps the new LLM API fields out of the legacy pybind KV cache config, and adds a guard for partially constructed KV cache managers when constructor validation fails.

Test Coverage

• Added LLM API validation coverage for KvCacheConfig.pool_ratio and KvCacheConfig.avg_seq_len.
• Added KV cache manager v2 coverage for explicit initial_pool_ratio overriding typical-step and constraints.
• Added DSV4 cache manager coverage for pool_ratio, avg_seq_len, and avg_seq_len > max_seq_len validation.
• Clean GB200 build on gb-nvl-081-compute02 with CUDA_ARCHITECTURES=100-real.
• Editable install in dsv4-pool-ratio-knob-jenkins-aarch64-jiaganc; TensorRT-LLM import reports 1.3.0rc15.
• Focused pytest run: 10 passed.
• git diff --check HEAD passed.
• python3 -m py_compile passed for changed Python files and tests.

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.

[jiaganc]

/bot run

[tensorrt-cicd]

PR_Github #50502 [ run ] triggered by Bot. Commit: 4499c60 Link to invocation

[lancelly]

LGTM, thanks!

[tensorrt-cicd]

PR_Github #50502 [ run ] completed with state SUCCESS. Commit: 4499c60
/LLM/main/L0_MergeRequest_PR pipeline #40010 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

[jiaganc]

/bot run

[tensorrt-cicd]

PR_Github #50526 [ run ] triggered by Bot. Commit: ef8d286 Link to invocation

[tensorrt-cicd]

PR_Github #50526 [ run ] completed with state SUCCESS. Commit: ef8d286
/LLM/main/L0_MergeRequest_PR pipeline #40033 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

[peihu-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #50591 [ run ] triggered by Bot. Commit: ef8d286 Link to invocation

[tensorrt-cicd]

PR_Github #50591 [ run ] completed with state SUCCESS. Commit: ef8d286
/LLM/main/L0_MergeRequest_PR pipeline #40089 completed with status: 'SUCCESS'

CI Report

Link to invocation