[https://nvbugs/6240561][fix] Autodeploy fix the deepseek accuracy drop

[nvchenghaoz]

Summary by CodeRabbit

• Tests
  • Improved validation of rotary embeddings with enhanced test assertions.

Description

Test Coverage

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)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• 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.

[coderabbitai]

📝 Walkthrough

Walkthrough

RoPE inverse-frequency and YaRN interpolation computation in the config fallback path is refactored: inverse frequencies now derive from full-dimension indices, YaRN scaling denominator computation is unified, and interpolation uses explicit mask-based blending between frequency terms with test validation added.

Changes

RoPE YaRN Computation Update

Layer / File(s)	Summary
RoPE inverse frequency and YaRN scaling setup tensorrt_llm/_torch/auto_deploy/transform/library/fuse_rope_mla.py	Inverse frequency is computed from even-dimension indices across full qk_rope_head_dim; YaRN denominator calculation always invokes _yarn_get_mscale() without conditional fallback.
YaRN interpolation mask-based blending tensorrt_llm/_torch/auto_deploy/transform/library/fuse_rope_mla.py, tests/unittest/auto_deploy/singlegpu/models/test_deepseek_custom.py	YaRN interpolation replaces smooth-step with inv_freq_mask-based explicit blending between freq_inter and freq_extra; test adds exact-match assertion before approximate comparison.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Suggested reviewers

• govind-ramnarayan
• bmarimuthu-nv
• MrGeva

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The PR description contains only the template with placeholders unfilled; no actual description, test coverage details, or explanation of changes is provided.	Fill in the Description section explaining the RoPE/YaRN computation changes and their purpose, add Test Coverage section listing relevant tests, and ensure the PR title follows the format with an appropriate ticket/issue reference.

✅ Passed checks (3 passed)

Check name	Status	Explanation
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.
Title check	✅ Passed	The title references the NVBugs ticket and fix type correctly, and specifically mentions 'deepseek accuracy drop' which aligns with the code changes targeting RoPE/YaRN rotary embedding fixes.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/unittest/auto_deploy/singlegpu/models/test_deepseek_custom.py`:
- Around line 192-194: The test currently uses torch.equal(actual, expected)
which enforces bitwise equality between actual (from
_compute_rotary_cos_sin_from_config(Factory()).cpu()) and expected (CPU),
causing cross-backend flakiness; remove that torch.equal check and rely only on
the tolerance-based assertion torch.testing.assert_close(actual, expected,
atol=3e-7, rtol=1e-4), or alternatively compute expected on the same backend as
actual by calling .to(actual.device) before comparing if you need strict
regression. Ensure changes target the assertions around
_compute_rotary_cos_sin_from_config, Factory, actual, and expected in this test.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 229e0bf0-ee11-4c7c-8b56-4765f60520da

📥 Commits

Reviewing files that changed from the base of the PR and between 74d7c3a and 3f40f8e.

📒 Files selected for processing (2)

• tensorrt_llm/_torch/auto_deploy/transform/library/fuse_rope_mla.py
• tests/unittest/auto_deploy/singlegpu/models/test_deepseek_custom.py

[taylor-yb-lee]

Issue link : https://jirasw.nvidia.com/browse/TRTLLM-13054

[nvchenghaoz]

/bot run --stage-list "A10-Build_Docs, A10-PackageSanityCheck-PY310-UB2204, A100X-PackageSanityCheck-PY312-UB2404, A30-AutoDeploy-1, H100_PCIe-AutoDeploy-1, DGX_B200-AutoDeploy-1, A100X-PyTorch-1, DGX_H100-4_GPUs-AutoDeploy-1, DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-Post-Merge-1, DGX_B200-8_GPUs-AutoDeploy-Post-Merge-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #51151 [ run ] triggered by Bot. Commit: 3f40f8e Link to invocation

[tensorrt-cicd]

PR_Github #51151 [ run ] completed with state SUCCESS. Commit: 3f40f8e
/LLM/main/L0_MergeRequest_PR pipeline #40585 (Partly Tested) 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