Update RAG metrics

[jmsevin]

Description

In order to evaluate future evolutions (choice of new LLM, hybrid search feature in Qdrant...), we needed to update the RAG metrics script.

Why?

Ragas library evolved quite a lot, so we had to update the metrics script to reflect those changes. Moreover, we decided to focus on metrics which don't need groundtruth data, as the way we synthesized them the first time may be not relevant for our current usecase. Finally, we added an option in the script to compute RAG metrics without context (i.e. when the LLM doesn't get resources from WeLearn database).

How?

Five RAG metrics are computed now:

• Context Precision (w/o reference)
• Response Relevancy
• Faithfulness
• Context Relevance
• Response Groundedness

The script is launched through the command line python rag-metrics.py. Three options can be added:

• --all_corpus to aggregate WeLearn resources by language and provide one context only (per language) to the LLM
• --reranking to rerank WeLearn resources in order to improve diversity
• --vanilla to compute RAG metrics without any context (i.e. WeLearn resources) for the LLM

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)

Checklist:

• My code follows the code style of this project.
• My code is tested.
• I have updated the documentation accordingly.

[Copilot]

Pull Request Overview

This PR updates the RAG metrics script and related chat service functions to align with the latest library changes and improve error handling.

• Added try/except blocks and enhanced logging in the chat functions.
• Updated dependency versions and added new dependencies in pyproject.toml to support evolved libraries.

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

File	Description
src/app/services/abst_chat.py	Improved error handling in chat calls and updated model strings.
pyproject.toml	Updated dependency versions for ragas and added new dependencies.

Comments suppressed due to low confidence (1)

src/app/services/abst_chat.py:595

• The print statement in the chat_schema function may be left over from debugging. Consider replacing it with a logging statement or removing it to avoid unintended console output in production.

print(completion)