[WIP] ONNX Model Quantization

[ethanreidel]

Code Pull Requests

Please provide the following:

• a clear explanation of what your code does Expands on @saad-palapa and @skanjila implementation of the ONNX exporter. Adds ability to quantize ONNX model to uint8 weights. Be aware that this PR does add 3 new .onnx model files strictly for testing based on squeezenet architecture.
• if applicable, a reference to an issue Addition to this issue: https://github.com/ludwig-ai/ludwig/issues/3659
• a reproducible test for your PR (code, config and data sample) tests->ludwig->model_export->test_onnx_exporter.py (tests->ludwig->model_export->sampleonnxmodel contains test onnx models)

Other issues: Currently test_quantize_onnx_model is passing but with a lot of warnings. Quantization parameters for tensor:"x" not specified Quantization parameters for tensor:"/features/features.2/MaxPool_output_0" not specified Continued...

Thanks for looking at this PR!

[ethanreidel]

Noting: need to add onnxruntime to requirements

[github-actions[bot]]

Unit Test Results

  6 files  ±  0  6 suites  ±0   13s :stopwatch: - 13m 54s   6 tests  -   6  0 :heavy_check_mark:  -   9    5 :zzz: +2  0 :x: ±0  1 :fire: +1  27 runs   - 33  0 :heavy_check_mark:  - 42  21 :zzz: +3  0 :x: ±0  6 :fire: +6 

For more details on these errors, see this check.

Results for commit fcc0abc2. ± Comparison against base commit 7e8fc14d.

This pull request removes 12 and adds 6 tests. Note that renamed tests count towards both.

``` tests.regression_tests.benchmark.test_model_performance ‑ test_performance[adult_census_income.ecd.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[adult_census_income.gbm.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[ames_housing.ecd.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[ames_housing.gbm.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[mercedes_benz_greener.ecd.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[mercedes_benz_greener.gbm.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[sarcos.ecd.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[sarcos.gbm.yaml] tests.regression_tests.model.test_old_models ‑ test_model_loaded_from_old_config_prediction_works tests.regression_tests.model.test_old_models ‑ test_predict_deprecated_model[respiratory] … ``` ``` tests.ludwig.automl.test_base_config tests.ludwig.automl.test_utils tests.ludwig.backend.test_ray tests.ludwig.benchmarking.test_profiler tests.ludwig.data.test_ray_data tests.ludwig.model_export.test_onnx_exporter ```

This pull request removes 3 skipped tests and adds 5 skipped tests. Note that renamed tests count towards both.

``` tests.regression_tests.benchmark.test_model_performance ‑ test_performance[ames_housing.ecd.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[mercedes_benz_greener.ecd.yaml] tests.regression_tests.benchmark.test_model_performance ‑ test_performance[sarcos.ecd.yaml] ``` ``` tests.ludwig.automl.test_base_config tests.ludwig.automl.test_utils tests.ludwig.backend.test_ray tests.ludwig.benchmarking.test_profiler tests.ludwig.data.test_ray_data ```