Quantization tool: support float 8 with MatMul, support float 16 weights #18043
Conversation
This file contains 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
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
yufenglee
reviewed
Jan 9, 2024
yufenglee
reviewed
Jan 9, 2024
yufenglee
approved these changes
Jan 11, 2024
There was a problem hiding this comment.
Please run the E2E resnet50 model to double check: https://github.com/microsoft/onnxruntime-inference-examples/tree/main/quantization/image_classification/cpu
adrianlizarraga
added a commit
that referenced
this pull request
Jan 13, 2024
…ta types (#19114) ### Description - Updates `get_qnn_qdq_config()` to use new scale/zp np.array data types. - Adds missing unit test to help prevent future regression. ### Motivation and Context #18043 changed the usage of `extra_options["TensorQuantizationOverrides"]`. We need to update its use in quantization/execution_providers/qnn/quant_config.py
mszhanyi
pushed a commit
that referenced
this pull request
Jan 15, 2024
…hts (#18043) ### Description Whenever a node QuantizeLinear or DequantizeLinear, the type of the weights before being quantize must be known to create the scale with the expected type. Another option would be to add many operator CastLike but that would push the burden to onnxruntime optimizer. The PR tries to avoid changing the signature. To do so, it modified the scale computation to use a numpy array to store the result and not a python float. The numpy array must be of the same type than the weights to quantize. The PR adds many `assert` to check the type of the scale is not a python type or a float64. This was added to make sure all the code follows the same logic. These lines were kept for the first review. DequantizeLinear, QuantizeLinear cannot be tested with onnx==1.15. PR onnx/onnx#5709 is missing to fix shape inference. PR onnx/onnx#5473) is missing to support QLinearMatMul with float 16. That explains why some tests are disabled with float 16. ### Motivation and Context The current quantization tool assumes every weight is float 32. For large models such as LLAMA, it is usually float 16. The quantization needs to quantize such weights.
mszhanyi
pushed a commit
that referenced
this pull request
Jan 15, 2024
…ta types (#19114) ### Description - Updates `get_qnn_qdq_config()` to use new scale/zp np.array data types. - Adds missing unit test to help prevent future regression. ### Motivation and Context #18043 changed the usage of `extra_options["TensorQuantizationOverrides"]`. We need to update its use in quantization/execution_providers/qnn/quant_config.py
5 tasks
guotuofeng
pushed a commit
to microsoft/Olive
that referenced
this pull request
Jan 17, 2024
## Describe your changes PR microsoft/onnxruntime#18043 (onnxruntime) extends onnxruntime quantization tools to support float16 weights. To do so, it enforces scale and zerop_point to be strongly typed (as `numpy.array(single_value, dtype=dtype)`). scale type should always be the weight type, and zero_point type the quantized weight type. That convention is checked all along the quantization tools to make sure there is loss of information. This change was made to avoid adding new arguments in many functions to carry this information. ## Checklist before requesting a review - [ ] Add unit tests for this change. - [ ] Make sure all tests can pass. - [ ] Update documents if necessary. - [ ] Lint and apply fixes to your code by running `lintrunner -a` - [ ] Is this a user-facing change? If yes, give a description of this change to be included in the release notes. ## (Optional) Issue link
xadupre
added a commit
to xadupre/onnxruntime
that referenced
this pull request
Jan 17, 2024
xadupre
added a commit
that referenced
this pull request
Jan 17, 2024
…19182) ### Description Extends the code coverage to Entroy, Histogram and Distribution calibration method, fix bugs while doing it. ### Motivation and Context Bugs detected in [Olive](https://github.com/microsoft/OLive).
YUNQIUGUO
pushed a commit
that referenced
this pull request
Jan 23, 2024
…19182) ### Description Extends the code coverage to Entroy, Histogram and Distribution calibration method, fix bugs while doing it. ### Motivation and Context Bugs detected in [Olive](https://github.com/microsoft/OLive).
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.
Description
Whenever a node QuantizeLinear or DequantizeLinear, the type of the weights before being quantize must be known to create the scale with the expected type. Another option would be to add many operator CastLike but that would push the burden to onnxruntime optimizer.
The PR tries to avoid changing the signature. To do so, it modified the scale computation to use a numpy array to store the result and not a python float. The numpy array must be of the same type than the weights to quantize.
The PR adds many
assertto check the type of the scale is not a python type or a float64. This was added to make sure all the code follows the same logic. These lines were kept for the first review.DequantizeLinear, QuantizeLinear cannot be tested with onnx==1.15. PR onnx/onnx#5709 is missing to fix shape inference. PR onnx/onnx#5473) is missing to support QLinearMatMul with float 16. That explains why some tests are disabled with float 16.
Motivation and Context
The current quantization tool assumes every weight is float 32. For large models such as LLAMA, it is usually float 16. The quantization needs to quantize such weights.