Razvan Dinu

@patriciapampanelli: can you provide more guidance on how to keep the LLM always responding in a different language? Was the "return to English" an issue in the tests you did...

Thanks for pointing this @joeywhelan! Indeed, we should update the docs. You have to declare a parameter for your action which has exactly the same name as the one you...

Hi @donatoaz ! This is a good question. Conceptually, what you want is a way to also get some "additional context" out of the generation. As a parallel, on the...

Thanks for following up on this @HanchenXiong. We do have support for this now: https://github.com/NVIDIA/NeMo-Guardrails/blob/develop/docs/user_guides/advanced/generation-options.md#output-variables. It was added in [0.8.0](https://github.com/NVIDIA/NeMo-Guardrails/blob/develop/CHANGELOG.md#080---2024-02-28).

Thanks for following up @shimada-r. The "stop tokens" fix did not make it to `0.7.0`/`0.7.1`. The PR https://github.com/NVIDIA/NeMo-Guardrails/pull/293 is in review now and should make it to `0.8.0`. Let's retest...

Not yet, but it's a feature that is already on our short list as it's been requested multiple times. Right now you can do a hack and make a convention...

Thanks for the follow up @Serbernari! Yes, it is now possible. It appears this feature is not documented. There's a brief mention in the docstring for the `generate_async` function: https://github.com/NVIDIA/NeMo-Guardrails/blob/develop/nemoguardrails/rails/llm/llmrails.py#L491....

@nashugame, can you provide more details? With such a huge difference (from 2-5 seconds to 5-10 mins) it doesn't sound like a performance issue, but rather a configuration issue.

@trebedea / @prasoonvarshney : can have a look at this?

@nairajay2k ! The bug should be fixed now. You can also install `nemoguardrails==0.8.0` directly from PyPI.