[6034518] Downgrade TRT support for remote autotuning in Autotune from 10.16 to 10.15

[gcunhase]

What does this PR do?

Type of change: Bug fix

Remote autotuning is supported in TensorRT from version 10.15, but fails with Autotune as it's checking for 10.16+. This PR fixes that check and updates documentation accordingly.

Usage

# Add a code snippet demonstrating how to use this

Testing

See bug 6034518.

Before your PR is "Ready for review"

• Is this change backward compatible?: ✅
• If you copied code from any other sources or added a new PIP dependency, did you follow guidance in CONTRIBUTING.md: N/A
• Did you write any new necessary tests?: N/A
• Did you update Changelog?: ✅

Summary by CodeRabbit

• Documentation

  • Added Remote Autotuning guide for TensorRT 10.15+ with CLI invocation examples and explicit requirements.
  • Updated examples and command documentation for remote autotuning configuration.

• Updates

  • Reduced TensorRT minimum version requirement for remote autotuning from 10.16 to 10.15.
  • Clarified CLI help text for autotuning arguments.

[coderabbitai]

📝 Walkthrough

Walkthrough

Updated TensorRT minimum version requirement for remote autotuning from 10.16 to 10.15 across changelog, documentation, examples, CLI help text, and version checks.

Changes

Cohort / File(s)	Summary
Changelog & Documentation CHANGELOG.rst, docs/source/guides/9_autotune.rst, examples/onnx_ptq/autotune/README.md	Updated TensorRT version from 10.16 to 10.15 for remote autotuning feature. Added new documentation section describing remote autotuning with safety mode, requirements, and CLI invocation example with --safe and --skipInference flags.
Implementation & CLI modelopt/onnx/quantization/__main__.py, modelopt/onnx/quantization/autotune/benchmark.py	Updated CLI help text to clarify --autotune_trtexec_args behavior and relevance to trtexec workflow. Changed version gate in remote autotuning check from 10.16 to 10.15 with corresponding log message updates.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately and concisely summarizes the main change: downgrading TRT version requirement for remote autotuning from 10.16 to 10.15.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Security Anti-Patterns	✅ Passed	No security anti-patterns found: no unsafe torch.load, numpy.load, trust_remote_code, eval/exec, or # nosec comments detected.

_{✏️ 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]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

modelopt/onnx/quantization/autotune/benchmark.py (1)

209-228: ⚠️ Potential issue | 🔴 Critical

Remove early return so remote autotuning args are actually applied.

Line 218 returns from __init__ before Line 227 extends self._base_cmd, so the --remoteAutoTuningConfig/--safe args never make it into the executed command when TRT >= 10.15. This breaks remote autotuning despite passing the version gate.

💡 Proposed fix

-        trtexec_args = self.trtexec_args or []
+        trtexec_args = list(self.trtexec_args or [])
         has_remote_config = any("--remoteAutoTuningConfig" in arg for arg in trtexec_args)

         if has_remote_config:
             try:
                 _check_for_tensorrt(min_version="10.15")
                 self.logger.debug("TensorRT Python API version >= 10.15 detected")
                 if "--safe" not in trtexec_args:
                     self.logger.warning(
                         "Remote autotuning requires '--safe' to be set. Adding it to trtexec arguments."
                     )
-                    self.trtexec_args.append("--safe")
-                return
+                    trtexec_args.append("--safe")
             except ImportError:
                 self.logger.warning(
                     "Remote autotuning is not supported with TensorRT version < 10.15. "
                     "Removing --remoteAutoTuningConfig from trtexec arguments"
                 )
                 trtexec_args = [
                     arg for arg in trtexec_args if "--remoteAutoTuningConfig" not in arg
                 ]
         self._base_cmd.extend(trtexec_args)

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@modelopt/onnx/quantization/autotune/benchmark.py` around lines 209 - 228, The
constructor's remote-autotuning branch returns early so trtexec args like
"--safe" or "--remoteAutoTuningConfig" never get appended to the final command;
remove the early return in the block that calls
_check_for_tensorrt(min_version="10.15") and instead ensure any modifications to
self.trtexec_args (and filtered local trtexec_args) are followed by extending
self._base_cmd with those trtexec_args so the arguments are applied; locate the
logic around _check_for_tensorrt, self.trtexec_args, trtexec_args and
self._base_cmd in the __init__ of the Benchmark class and adjust flow so the
successful TensorRT path does not exit before
self._base_cmd.extend(trtexec_args) executes.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Outside diff comments:
In `@modelopt/onnx/quantization/autotune/benchmark.py`:
- Around line 209-228: The constructor's remote-autotuning branch returns early
so trtexec args like "--safe" or "--remoteAutoTuningConfig" never get appended
to the final command; remove the early return in the block that calls
_check_for_tensorrt(min_version="10.15") and instead ensure any modifications to
self.trtexec_args (and filtered local trtexec_args) are followed by extending
self._base_cmd with those trtexec_args so the arguments are applied; locate the
logic around _check_for_tensorrt, self.trtexec_args, trtexec_args and
self._base_cmd in the __init__ of the Benchmark class and adjust flow so the
successful TensorRT path does not exit before
self._base_cmd.extend(trtexec_args) executes.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: aca1a5f5-c803-49d9-8889-37357600cffb

📥 Commits

Reviewing files that changed from the base of the PR and between b6c6ec3 and 2097254.

📒 Files selected for processing (5)

• CHANGELOG.rst
• docs/source/guides/9_autotune.rst
• examples/onnx_ptq/autotune/README.md
• modelopt/onnx/quantization/__main__.py
• modelopt/onnx/quantization/autotune/benchmark.py

[ajrasane]

PR #1259 Review: Downgrade TRT support for remote autotuning from 10.16 to 10.15

Verdict: Request Changes — core fix is correct, but docs have formatting issues and there's a code/doc inconsistency.

---

1. RST formatting errors — docs/source/guides/9_autotune.rst

• Missing blank line after .. code-block:: bash — RST requires a blank line between the directive and content. All other code blocks in this file follow this pattern.
• Wrong indentation — content uses 2-space indent but every other code block in the file uses 3-space indent. This will prevent proper rendering.
• Stray Markdown fence — there's a ``` closing fence after the code block that doesn't belong in RST. Will render as literal text.
• Leading space on " Other TensorRT benchmark options..." line — creates an unintended blockquote.

2. --skipInference inconsistency between docs and code

Docs (both RST and README) now state --safe --skipInference must be enabled, but benchmark.py:211-215 only auto-adds --safe — no check or auto-add for --skipInference:

  if "--safe" not in trtexec_args:                                                                                                                 
      self.logger.warning(...)                                                                                                      
      self.trtexec_args.append("--safe")                                                            
  # No equivalent for --skipInference

Either:

• Add a similar auto-add for --skipInference in the code (preferred for consistency), or
• Clarify in docs that --skipInference is recommended but not enforced

3. Minor: main.py help text

The updated help says "Only relevant with the 'trtexec' workflow enabled" — would be clearer as "Only relevant when --use_trtexec is set" since that's the actual flag name.

What looks good

• Version downgrade 10.16 → 10.15 in benchmark.py — correct and consistent across all 3 occurrences
• CHANGELOG.rst entry — properly placed
• README.md updates — version numbers and --skipInference addition are consistent

[codecov]

Codecov Report

❌ Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 77.45%. Comparing base (b6c6ec3) to head (2097254).

Files with missing lines	Patch %	Lines
modelopt/onnx/quantization/autotune/benchmark.py	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1259      +/-   ##
==========================================
+ Coverage   76.90%   77.45%   +0.54%     
==========================================
  Files         350      350              
  Lines       40524    40524              
==========================================
+ Hits        31166    31388     +222     
+ Misses       9358     9136     -222     

Flag	Coverage Δ
examples	43.81% <0.00%> (+1.18%)	⬆️
gpu	57.43% <0.00%> (-0.10%)	⬇️
unit	55.59% <0.00%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.