[None][fix] Warn on experimental DeepSeek-V4 base checkpoints

[lfr-0531]

@coderabbitai summary

Description

DeepSeek-V4 Base checkpoints currently go through the same config path as the better supported DeepSeek-V4 Instruct checkpoints, but Base support is still experimental. The Base checkpoints in /raid/model can be distinguished from Instruct checkpoints by routed MoE safetensors metadata: Instruct expert tensors use the existing MXFP4/NVFP4 layouts, while Base expert tensors are not MXFP4 or NVFP4.

This PR adds a small DeepSeek-V4 Base checkpoint detector that reuses the routed MoE tensor metadata check and emits a warning when users load a Base checkpoint. The warning tells users that Base support is experimental and recommends using a DeepSeek-V4 Instruct checkpoint.

Test Coverage

Added unit coverage for DeepSeek-V4 routed expert metadata detection:

• MXFP4 metadata is detected as a non-Base checkpoint.
• NVFP4 metadata is detected as a non-Base checkpoint.
• FP8 routed expert metadata is detected as a Base checkpoint.

Validation run locally:

• PATH=/usr/bin:$PATH pre-commit run --files tensorrt_llm/_torch/model_config.py tests/unittest/_torch/test_model_config.py
• /usr/bin/python3.12 -m py_compile tensorrt_llm/_torch/model_config.py tests/unittest/_torch/test_model_config.py

Not run successfully:

• PYTHONPATH=$PWD pytest tests/unittest/_torch/test_model_config.py -k deepseek_v4_base_checkpoint_detection fails during test bootstrap because the local environment is missing nvtx.

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 --disable-fail-fast

[tensorrt-cicd]

PR_Github #49153 [ run ] triggered by Bot. Commit: 83d0dd9 Link to invocation

[tensorrt-cicd]

PR_Github #49153 [ run ] completed with state SUCCESS. Commit: 83d0dd9
/LLM/main/L0_MergeRequest_PR pipeline #38834 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

[lfr-0531]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #49296 [ run ] triggered by Bot. Commit: aaadd6f Link to invocation

[tensorrt-cicd]

PR_Github #49296 [ run ] completed with state SUCCESS. Commit: aaadd6f
/LLM/main/L0_MergeRequest_PR pipeline #38960 completed with status: 'SUCCESS'

CI Report

Link to invocation