[OMNIML-2244] Add E2E example for mixed precision quantization and ONNX export #656
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #656 +/- ##
==========================================
- Coverage 74.50% 74.46% -0.05%
==========================================
Files 183 183
Lines 18400 18415 +15
==========================================
+ Hits 13709 13712 +3
- Misses 4691 4703 +12 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
…NX export Signed-off-by: ajrasane <[email protected]>
Signed-off-by: ajrasane <[email protected]>
Signed-off-by: ajrasane <[email protected]>
ajrasane
force-pushed
the
ajrasane/mixed_precision_e2e
branch
from
23e00a2 to
f403387
Compare
cjluo-nv
reviewed
Dec 8, 2025
cjluo-nv
reviewed
Dec 8, 2025
cjluo-nv
reviewed
Dec 8, 2025
Contributor
|
We should add perf and accuracy numbers for the baseline and quantized models in the README file as well. |
Contributor
|
A basic query: Is onnx_ptq the right place of "PyTorch PTQ => ONNX export" examples? I was under the impression that onnx_ptq exemplifies PTQ techniques for input ONNX models. |
Contributor
Right now, |
Signed-off-by: ajrasane <[email protected]>
ajrasane
force-pushed
the
ajrasane/mixed_precision_e2e
branch
from
0211848 to
9758f0c
Compare
Collaborator
|
You'll need to change test from |
ajrasane
force-pushed
the
ajrasane/mixed_precision_e2e
branch
from
9758f0c to
ee0d90f
Compare
gcunhase
reviewed
Dec 10, 2025
| | :--- | :---: | :---: | | ||
| | Torch autocast (FP16) | 85.11% | 97.53% | | ||
| | NVFP4 Quantized | 84.558% | 97.36% | | ||
| | Auto Quantized (FP8 + NVFP4, 4.78 effective bits) | 84.726% | 97.434% | |
Contributor
There was a problem hiding this comment.
NVFP4 Quantized is NVFP4 + FP16-AutoCast? Same question for Auto Quantized.
gcunhase
approved these changes
Dec 10, 2025
Contributor
gcunhase
left a comment
gcunhase
left a comment
There was a problem hiding this comment.
Thanks for dividing the examples into 2 folders.
I posted a couple more comments and we'd still need to add the perf numbers to show the accuracy-runtime trade-offs.
Approving for now.
kevalmorabia97
approved these changes
Dec 11, 2025
ajrasane
force-pushed
the
ajrasane/mixed_precision_e2e
branch
from
ee0d90f to
11de7ce
Compare
Signed-off-by: ajrasane <[email protected]>
ajrasane
force-pushed
the
ajrasane/mixed_precision_e2e
branch
from
11de7ce to
6c9f7b1
Compare
cjluo-nv
reviewed
Dec 12, 2025
| | :--- | :---: | :---: | | ||
| | Torch autocast (FP16) | 85.11% | 97.53% | | ||
| | NVFP4 Quantized | 84.558% | 97.36% | | ||
| | Auto Quantized (FP8 + NVFP4, 4.78 effective bits) | 84.726% | 97.434% | |
Collaborator
There was a problem hiding this comment.
can we just say 4.8 effective bits as this is what you set in the command above?
cjluo-nv
approved these changes
Dec 12, 2025
b7r6
pushed a commit
to weyl-ai/Model-Optimizer
that referenced
this pull request
Dec 18, 2025
…NX export (NVIDIA#656) ## What does this PR do? **Type of change:** New Feature **Overview:** - Enable ONNX export for auto quantized models - Update documentation and changelog ## Usage <!-- You can potentially add a usage example below. --> ``` python torch_quant_to_onnx.py --quantize_mode=auto \ --onnx_save_path=./vit_base_patch16_224.nvfp4_fp8.onnx \ --calibration_data_size 64 \ --auto_quantization_formats NVFP4_AWQ_LITE_CFG FP8_DEFAULT_CFG \ --batch_size 128 ``` ## Testing <!-- Mention how have you tested your change if applicable. --> ``` python evaluate.py --onnx_path=vit_base_patch16_224.nvfp4_fp8.onnx \ --model_name=vit_base_patch16_224 \ --results_path=./results.txt \ --batch_size 128 ``` Accuracy results ``` The top1 accuracy of the model is 84.15% The top5 accuracy of the model is 97.396% ``` Reference accuracy for fp16 ``` The top1 accuracy of the model is 85.102% The top5 accuracy of the model is 97.526% ``` ## Before your PR is "*Ready for review*" <!-- If you haven't finished some of the above items you can still open `Draft` PR. --> - **Make sure you read and follow [Contributor guidelines](https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/CONTRIBUTING.md)** and your commits are signed. - **Is this change backward compatible?**: Yes - **Did you write any new necessary tests?**: No - **Did you add or update any necessary documentation?**: Yes - **Did you update [Changelog](https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/CHANGELOG.rst)?**: Yes <!--- Only for new features, API changes, critical bug fixes or bw breaking changes. --> --------- Signed-off-by: ajrasane <[email protected]>
soodoshll
pushed a commit
to soodoshll/TensorRT-Model-Optimizer
that referenced
this pull request
Dec 18, 2025
…NX export (NVIDIA#656) ## What does this PR do? **Type of change:** New Feature **Overview:** - Enable ONNX export for auto quantized models - Update documentation and changelog ## Usage <!-- You can potentially add a usage example below. --> ``` python torch_quant_to_onnx.py --quantize_mode=auto \ --onnx_save_path=./vit_base_patch16_224.nvfp4_fp8.onnx \ --calibration_data_size 64 \ --auto_quantization_formats NVFP4_AWQ_LITE_CFG FP8_DEFAULT_CFG \ --batch_size 128 ``` ## Testing <!-- Mention how have you tested your change if applicable. --> ``` python evaluate.py --onnx_path=vit_base_patch16_224.nvfp4_fp8.onnx \ --model_name=vit_base_patch16_224 \ --results_path=./results.txt \ --batch_size 128 ``` Accuracy results ``` The top1 accuracy of the model is 84.15% The top5 accuracy of the model is 97.396% ``` Reference accuracy for fp16 ``` The top1 accuracy of the model is 85.102% The top5 accuracy of the model is 97.526% ``` ## Before your PR is "*Ready for review*" <!-- If you haven't finished some of the above items you can still open `Draft` PR. --> - **Make sure you read and follow [Contributor guidelines](https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/CONTRIBUTING.md)** and your commits are signed. - **Is this change backward compatible?**: Yes - **Did you write any new necessary tests?**: No - **Did you add or update any necessary documentation?**: Yes - **Did you update [Changelog](https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/CHANGELOG.rst)?**: Yes <!--- Only for new features, API changes, critical bug fixes or bw breaking changes. --> --------- Signed-off-by: ajrasane <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Type of change:
New Feature
Overview:
Usage
Testing
Accuracy results
Reference accuracy for fp16
Before your PR is "Ready for review"