shub-kris commented on code in PR #23554: URL: https://github.com/apache/beam/pull/23554#discussion_r991874721
########## website/www/site/content/en/documentation/ml/runinference-metrics.md: ########## @@ -0,0 +1,103 @@ +--- +title: "RunInference Metrics" +--- +<!-- +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + +http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +--> + +# RunInference Metrics Example + +The main purpose of the example is to demonstrate and explain different metrics that are available when using [RunInference](https://beam.apache.org/documentation/transforms/python/elementwise/runinference/) for doing inference using a machine learning model. We use a pipeline that reads a list of sentences, tokeinze the text, uses a Transformer based model `distilbert-base-uncased-finetuned-sst-2-english` for classifies the texts into two different classes using `RunInference`. + +We showcase different RunInference metrics when the pipeline is executed using Dataflow Runner on CPU and GPU. The full example code can be found [here](https://github.com/apache/beam/tree/master/sdks/python/apache_beam/examples/inference/runinference_metrics/). + + +The file structure for entire pipeline is: + + runinference_metrics/ + ├── pipeline/ + │ ├── __init__.py + │ ├── options.py + │ └── transformations.py + ├── __init__.py + ├── config.py + ├── main.py + └── setup.py + +`pipeline/transormations.py` contains the code for `beam.DoFn` and additional functions that are used for pipeline + +`pipeline/options.py` contains the pipeline options to configure the Dataflow pipeline + +`config.py` defines some variables like GCP PROJECT_ID, NUM_WORKERS that are used multiple times + +`setup.py` defines the packages/requirements for the pipeline to run + +`main.py` contains the pipeline code and some additional functions used for running the pipeline + + +### How to Run the Pipeline ? +First, make sure you have installed the required packages. One should have access to a Google Cloud Project and then correctly configure the GCP variables like `PROJECT_ID`, `REGION`, and others in `config.py`. For using Dataflow with GPU, all the necessary setup instructions are mentioned here: https://github.com/GoogleCloudPlatform/python-docs-samples/tree/main/dataflow/gpu-examples/pytorch-minimal. + + +1. Dataflow with CPU: `python main.py --mode cloud --device CPU` +2. Dataflow with GPU: `python main.py --mode cloud --device GPU` + +The pipeline can be broken down into few simple steps: +1. Create a list of texts to use it as an input using `beam.Create` +2. Tokenizing the text +3. Using RunInference to do inference +4. Postprocessing the output of RunInference + +The code snippet for the pipeline is: + +{{< highlight >}} + with beam.Pipeline(options=pipeline_options) as pipeline: + _ = ( + pipeline + | "Create inputs" >> beam.Create(inputs) + | "Tokenize" >> beam.ParDo(Tokenize(cfg.TOKENIZER_NAME)) + | "Inference" >> + RunInference(model_handler=KeyedModelHandler(model_handler)) + | "Decode Predictions" >> beam.ParDo(PostProcessor())) +{{< /highlight >}} + + +## RunInference Metrics + +As mentioned above, we benchmarked the performance of RunInference using Dataflow on both CPU and GPU. These metrics can be seen in the GCP UI and can also be printed using +{{< highlight >}} +metrics = pipeline.result.metrics().query(beam.metrics.MetricsFilter()) +{{< /highlight >}} + + +A snapshot of different metrics from GCP UI when using Dataflow on GPU: + +  Review Comment: It's a snapshot from the GCP UI. Right now it is hosted on drive and shared to everyone. -- 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]
