Feature/add mistral generator

[dimensi0n]

Mistral is an advanced language model that offers powerful and precise text generation capabilities. It has been specifically designed to follow precise instructions, reason, handle multi-turn conversations, and generate code. By integrating Mistral into Garak, we enable users to test this state-of-the-art model against various attacks and vulnerabilities, ensuring safer and more reliable use in real-world applications.

Verification

$ export MISTRAL_API_KEY=abcdefg123456789
$ python3 -m garak --model_type mistral --model_name mistral-small-latest --probes test.Blank

The default model name is [mistral-large-latest] (https://mistral.ai/fr/news/mistral-large-2407)

[github-actions]

DCO Assistant Lite bot All contributors have signed the DCO ✍️ ✅

[dimensi0n]

I have read the DCO Document and I hereby sign the DCO

[dimensi0n]

recheck

[jmartin-tech]

Looks like a good first pass, API keys and default values are part for the Configurable plugin implementation. I have offered a refactor that would move this validation and default to enable our yaml and json based configuration to work with this generator.

It would be helpful to have a few example generation responses as mocks for pytests added to validate this new generator code.

[dimensi0n]

How to add the mistral api key for the pytests in github actions ?

[jmartin-tech]

How to add the mistral api key for the pytests in github actions ?

Tests can exist that require a key for usage in integration testing however, these test are not run in CI.

Github actions will not use a live API key, tests should exists that mock the http response and intercept the API call.

[dimensi0n]

Here we go, we have a mock prompt test, pytests should be fine now.

[dimensi0n]

Thanks for the example, it was a hard path to find how to create the mock file but I found this on my way https://github.com/mistralai/client-python/blob/main/docs/models/chatcompletionresponse.md. Should be done with the last commit

[jmartin-tech]

@dimensi0n this is great progress, I noted a question about why the code is pushing a value into the environment that is explicitly passed to the client object and have requested some more work in the backoff handling.

[dimensi0n]

Up