Add module config for module_example and module_text_llm

[FelixTJDietrich]

Motivation and Context

We cant to test out multiple module configurations mostly for evaluation purposes, but it can also be used by the LMS to have some more control over the module. For my llm modules I want to be able to configure multiple approaches, multiple LLMs, and also prompts for the LLMs.

Another thing is that we might want to configure the module to be in debug mode or evaluation mode and send back debug data or evaluation data. For example for the LLMs, we would like to track the token usage, or the exact prompts used, etc.

Description

Small illustration

List of changes

• Add config_schema_provider decorator to athena (GET /config_schema endpoint for module sending back possible config options)
• Also proxy GET requests in assessment_module_manager to modules
• Add example config_schema_provider to module_example
• Fix error that made module_example not run
• Playground updates
  • Display /config_schema request in Module Requests as Get Config Schema
  • Display configs using react-jsonschema-forms in the BaseInfoHeader (it tries to render the schema if it is found, config component can be customized)
  • Send module config with request headers to Athena (if enabled)
• Add example of how to consume the config in module_example
• Update module_llm_text to support azure and non-azure credentials at the same time
• Add config_schema_provider to module_text_llm
  • Available models (LLMs)
  • Available approaches
  • Available prompts and their plain text (using monaco editor)
• Rework openai model config
• Use config in module_text_llm
  • Use model
  • Use approach
  • Use prompts
• Add emit_meta and get_meta to athena and include the emitted metadata with the request response

Steps for Testing

Prerequisites:
Update environment variables for module_text_llm or alternatively use module_example for testing, but it is less cool.

1. Do all requests in the playground without overriding the config
2. Do all requests in the playground with overriding the config for module_text_llm in the BaseInfoHeader
   • Setting debug to true sends back metadata (prompt & result) in the request feedback suggestions request
   • Prompt from the metadata should be the same of the config

Screenshots

Screenshot cut off....

[FelixTJDietrich]

The error you experienced with my module is still on my todo list. I will replace CSV parsing through JSON parsing and improve error handling so this does not happen anymore :S

[pal03377]

Nice changes! I only have larger problems in request_to_module.py left now. I also tested it locally again.

Also, I notice that endpoints.py is getting more and more complex (which is not really your fault). We might need to think of some refactoring in the future.

[pal03377]

Code looks good & I tested it again locally. Thanks for the contribution, very cool changes! I'll merge it right now