LLM-based example

[tectonia]

This pull request introduces a new LLM-based example - the invoice processing pipeline, including new configurations, dependencies, workflows, and documentation. Key changes focus on enabling invoice processing experiments, integrating Azure OpenAI services, and enhancing pipeline functionality.

Invoice Processing Pipeline:

• New CI Workflow: Added invoice_processing_ci_pipeline.yml to define a continuous integration workflow for invoice processing, supporting pull requests and workflow dispatch triggers.
• Pipeline Configurations: Updated config/config.yaml to include configurations for invoice processing pipelines (invoice_processing_pr and invoice_processing_dev) with specific compute cluster and dataset settings. [1] [2]
• Experiment Configuration: Added experiment_config.yaml to define parameters for data preparation, prediction, and scoring in invoice processing experiments.
• Predict Component: Created mlops/invoice_processing/components/predict.yml to define the predict component for the pipeline, integrating Azure OpenAI service inputs and outputs.

Dependency Updates:

• Requirements Files: Updated dependencies in .github/requirements/build_validation_requirements.txt and .github/requirements/execute_job_requirements.txt to include newer versions of mlflow, azure-ai-ml, and additional libraries like azureml-fsspec, Levenshtein, and python-retry. [1] [2]

Workflow Improvements:

• Build Validation Workflow: Enhanced build_validation_workflow.yml to include PYTHONPATH for improved test execution.
• Azure CLI Token Cache: Added a step to clear Azure CLI token cache in execute_shell_code/action.yml for better security and reliability.

Documentation Updates:

• Experiment Configuration Guide: Added docs/how-to/ConfigureExperiments.md to provide detailed instructions on configuring experiments, including .env file setup and pipeline configurations.
• Prompts and Strategies: Added docs/how-to/PromptsAndExtractionStrategies.md to document prompt creation and extraction strategies for invoice processing.

Data and Config Updates:

• Data Configuration: Updated config/data_config.json to include new datasets for invoice processing (invoice_processing_test and invoice_processing_test_gt).
• Config Utils: Enhanced mlops/common/config_utils.py to support loading experiment_config.yaml alongside the main configuration file.

These changes collectively enable robust support for the invoice processing pipeline, streamline workflows, and enhance documentation for easier onboarding and experimentation.