[None][fix] compare library versions with packaging.version, not strings

[lonexreb]

Description

Three version gates compared __version__ strings with >=, which is lexicographic, not numeric:

• tensorrt_llm/_torch/models/__init__.py — transformers.__version__ >= "4.45.1" (gates MllamaForConditionalGeneration registration)
• tensorrt_llm/tools/multimodal_builder.py — transformers.__version__ >= '4.53.0'
• tensorrt_llm/_torch/speculative/interface.py — flashinfer.__version__ >= "0.6.4"

Lexicographic comparison is wrong at version boundaries, e.g. "4.9.0" >= "4.45.1" is True (should be False) and "4.100.0" >= "4.53.0" is False (should be True).

Change

Use packaging.version.parse(...) for numeric comparison — the same pattern already used in _torch/modules/mamba/softplus.py and _torch/modules/fla/utils.py. Behavior is unchanged for currently-supported versions; this only corrects the edge cases.

Test

No dedicated test: these are import-time gates and the change is a standard packaging.version substitution with no behavior change for supported versions.

Summary by CodeRabbit

• Bug Fixes
  • Improved compatibility checks for Transformers and FlashInfer versions.
  • Ensured version comparisons correctly follow semantic versioning, including multi-digit releases.
  • Improved selection of multimodal attention behavior across supported Transformers versions.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: c103299c-35b9-423b-8f3d-40938c173d34

📥 Commits

Reviewing files that changed from the base of the PR and between b4e7808 and 3742e58.

📒 Files selected for processing (3)

• tensorrt_llm/_torch/models/__init__.py
• tensorrt_llm/_torch/speculative/interface.py
• tensorrt_llm/tools/multimodal_builder.py

---

📝 Walkthrough

Walkthrough

The pull request replaces direct string comparisons with packaging.version.parse() for Transformers and FlashInfer compatibility checks in model exports, speculative decoding, and multimodal builder initialization.

Changes

Semantic version gates

Layer / File(s)	Summary
Parsed compatibility checks tensorrt_llm/_torch/models/__init__.py, tensorrt_llm/_torch/speculative/interface.py, tensorrt_llm/tools/multimodal_builder.py	Version checks against Transformers 4.45.1, FlashInfer 0.6.4, and Transformers 4.53.0 now use semantic version parsing while preserving the existing conditional behavior.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title matches the PR’s main change and uses the repository’s required [None][fix] format.
Description check	✅ Passed	The description explains the bug, the fix, and test coverage; only the checklist section is not explicitly completed.
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

⚔️ Resolve merge conflicts

• Resolve merge conflict in branch fix/version-string-compare

---

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