[None][fix] Run DeepSeek V4 gate test on CUDA

[lfr-0531]

@coderabbitai summary

Description

The DeepSeek V4 gate unit test was added to the multi-GPU CI test list, where it exposed that test_deepseek_v4_gate_uses_fp32_reference_linear constructed the gate and input tensors on CPU.

DeepseekV4Gate.forward() calls trtllm::dsv3_router_gemm_op, which is a CUDA-only custom op. This caused the B200 multi-GPU stage to fail with a CPU backend dispatch error.

This PR updates the test to run the gate, hidden states, and reference weights on CUDA, and skips the test when CUDA is unavailable.

Test Coverage

• python3 -m pytest -q tests/unittest/_torch/modeling/test_modeling_deepseekv4.py::test_deepseek_v4_gate_uses_fp32_reference_linear
• python3 -m pytest -q tests/unittest/_torch/modeling/test_modeling_deepseekv4.py
• pre-commit run --files tests/unittest/_torch/modeling/test_modeling_deepseekv4.py

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.

[lfr-0531]

/bot run --add-multi-gpu-test

[tensorrt-cicd]

PR_Github #47482 [ run ] triggered by Bot. Commit: ab7217e Link to invocation