karthik-0306 opened a new pull request, #41389: URL: https://github.com/apache/beam/pull/41389
Fixes #41384 `OnnxModelHandlerNumpy` and `XGBoostModelHandler` didn't implement `update_model_path`, so when `RunInference` got a new `ModelMetadata` on `model_metadata_pcoll` (for example from `WatchFilePattern`), `load_model()` reloaded the original `model_uri` / `model_state` and the pipeline kept using the old model without an error. Both handlers now update the path, as the PyTorch, sklearn, TensorFlow and HuggingFace handlers already do. The update is on the `XGBoostModelHandler` base class, so the Numpy, Pandas, SciPy and Datatable handlers all get it. ONNX predictions now also set `model_id`, which was always `None`. Testing: - `test_pipeline_updates_model_from_side_input` starts on model a (y = 2x + 0.5) and sends model b (y = 5x + 0.5) on the side input. On master it gets `4.5` with `model_id=None`; with the fix it gets `10.5` with `model_id` set to model b. - The new `test_update_model_path` tests for ONNX and XGBoost fail on master and pass with the fix. - Three existing ONNX tests now expect `model_id` to be set. - `onnx_inference_test.py` (TensorFlow tests not run locally): 12 passed, 1 skipped. `xgboost_inference_test.py`, `sklearn_inference_test.py` and `base_test.py`: 177 passed, 4 skipped. - yapf and ruff are clean on the touched files. CHANGES.md updated. R: @jrmccluskey -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
