[6466805] Add Dynamo ONNX export support for quantized models

[ajrasane]

What does this PR do?

Type of change: new feature

• Adds Dynamo ONNX export support for FP8, INT8, INT4 AWQ, MXFP8, NVFP4, and mixed AutoQuant models with ONNX opset 21 or newer.
• Emits quantization nodes directly during normal Dynamo capture and provides get_dynamo_onnx_translation_table() for strict capture.
• Integrates strict translations and block-weight postprocessing into the ModelOpt ONNX helper.
• Preserves the legacy TorchScript exporter and existing defaults.
• Adds CLI options, documentation, changelog coverage, and CPU/GPU regression tests.

Usage

import torch

from modelopt.torch._deploy.utils import get_onnx_bytes_and_metadata
from modelopt.torch.quantization.export_onnx import get_dynamo_onnx_translation_table

onnx_bytes, metadata = get_onnx_bytes_and_metadata(
    model,
    (sample_input,),
    dynamo_export=True,
    onnx_opset=24,
)

exported_program = torch.export.export(model, (sample_input,), strict=True)
torch.onnx.export(
    exported_program,
    (),
    "model.onnx",
    dynamo=True,
    opset_version=24,
    custom_translation_table=get_dynamo_onnx_translation_table(),
)

Testing

• CUDA-hidden unit coverage passed across supported PyTorch 2.8–2.13 environments.
• Final MXFP8 metadata regression passed on PyTorch 2.8 and 2.13.
• GPU Dynamo exports passed for FP8, INT8, and MXFP8.
• NVFP4 opset 21/24 exports and TensorRT builds passed on Blackwell.
• The FP8 CLI smoke passed with --dynamo_export --onnx_opset=24 --trt_build.
• All changed-file pre-commit hooks and git diff --check passed.

Before your PR is "Ready for review"

Make sure you read and follow Contributor guidelines and your commits are signed (git commit -s -S).

Make sure you read and follow the Security Best Practices (e.g. avoiding hardcoded trust_remote_code=True, torch.load(..., weights_only=False), pickle, etc.).

• Is this change backward compatible?: ✅ — Dynamo export is opt-in; existing exporter and opset defaults are unchanged.
• 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?: ✅
• Did you update Changelog?: ✅

🤖 Generated by Codex (AI agent).

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

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

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://NVIDIA.github.io/Model-Optimizer/pr-preview/pr-2321/
Built to branch gh-pages at 2026-09-03 02:11 UTC. Preview will be ready when the GitHub Pages deployment is complete.