[None][fix] Align dense Qwen3.5-VL SSM cache dtype test with #16065 semantics

[moraxu]

Summary by CodeRabbit

• Tests
  • Updated validation coverage for automatic Mamba SSM KV-cache dtype selection.
  • Confirmed automatic allocation uses bfloat16, while explicit float32 configuration remains supported.

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.

[moraxu]

/bot run

[coderabbitai]

📝 Walkthrough

Walkthrough

The dense Qwen3.5-VL Mamba SSM cache dtype test now expects automatic resolution to torch.bfloat16, verifies the extracted cache parameters, and retains torch.float32 for explicit opt-in.

Changes

Qwen3.5-VL Mamba cache dtype validation

Layer / File(s)	Summary
Update cache dtype assertions tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py	The test now validates torch.bfloat16 for automatic cache dtype resolution and extracted parameters, while retaining the explicit float32 expectation.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#15249: Updates the Mamba cache dtype resolution logic referenced by these test expectations.
• NVIDIA/TensorRT-LLM#16065: Updates Mamba SSM cache dtype resolution and fallback behavior covered by this test.

Suggested reviewers: achartier

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description is mostly template text and lacks a real issue summary, solution details, and test coverage.	Replace the placeholder text with a short problem/solution summary and add concrete test coverage for the changed test.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title matches the change and follows the required [None][fix] format.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[coderabbitai]

🧹 Nitpick comments (1)

tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py (1)

160-163: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Consider asserting extract_mamba_kv_cache_params for the explicit float32 path too.

The auto path verifies both quant_config.mamba_ssm_cache_dtype and the downstream extract_mamba_kv_cache_params(...).mamba_ssm_cache_dtype (lines 153-158), but the explicit "float32" opt-in path only asserts quant_config.mamba_ssm_cache_dtype. For symmetry and to catch regressions in _coerce_torch_dtype handling of torch.float32 on the extraction side, consider adding the same extraction-path check here.

🧪 Proposed addition

     opt_in_config = ModelConfig(pretrained_config=config)
     validate_and_set_mamba_ssm_cache_dtype(opt_in_config, "float32")
     assert opt_in_config.quant_config.mamba_ssm_cache_dtype is torch.float32
+
+    opt_in_mamba_params = extract_mamba_kv_cache_params(
+        config.text_config,
+        quant_config=opt_in_config.quant_config,
+    )
+    assert opt_in_mamba_params.mamba_ssm_cache_dtype is torch.float32

As per path instructions for tests/**, coverage here is largely sufficient but this symmetric check would close a small gap.

🤖 Prompt for 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.

In `@tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py` around lines 160
- 163, Extend the explicit float32 opt-in test around
validate_and_set_mamba_ssm_cache_dtype to also call
extract_mamba_kv_cache_params and assert its mamba_ssm_cache_dtype is
torch.float32, matching the existing auto-path coverage and validating
downstream dtype coercion.

Source: Path instructions

🤖 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.

Nitpick comments:
In `@tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py`:
- Around line 160-163: Extend the explicit float32 opt-in test around
validate_and_set_mamba_ssm_cache_dtype to also call
extract_mamba_kv_cache_params and assert its mamba_ssm_cache_dtype is
torch.float32, matching the existing auto-path coverage and validating
downstream dtype coercion.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: a416c4de-da88-4d0b-bd4a-88184eed1ddc

📥 Commits

Reviewing files that changed from the base of the PR and between b4e7808 and ac278c5.

📒 Files selected for processing (1)

• tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py

[tensorrt-cicd]

PR_Github #58719 [ run ] triggered by Bot. Commit: ac278c5 Link to invocation

[tensorrt-cicd]

PR_Github #58719 [ run ] completed with state SUCCESS. Commit: ac278c5
/LLM/main/L0_MergeRequest_PR pipeline #47302 completed with status: 'SUCCESS'

CI Report

Link to invocation