Adds Groq support

[masegraye]

Groq is largely OpenAI compatible, with a few quirks.

• Adds first-class Groq support
• Includes searching in response for x_grok, for token usage information
• Adds proof-of-life test for all supported production models
• All API tests tagged live_groq
• ./scripts/groq_chat.sh lets you interactively test chatting with Groq models

NOTE: I was only able to test this code in isolation - not with anything that required the various keys in test_helper.exs be set.

[brainlid]

Groq support was added in #338

[masegraye]

@brainlid This was for Groq (https://groq.com/) not Grok (https://x.ai/api). Groq is a managed provider for open models that provides fast inference.

[masegraye]

@brainlid previously I was content to use this in my personal fork, but I find myself needing this again in an official capacity for work. Before I work to bring it back up to date with the latest langchain, I should ask: is this something you're likely to accept as a contribution?

[brainlid]

Hi @masegraye! Sorry, I didn't catch the service difference.

Yes, I'd accept this as a contribution. I'll reopen it. However, there are a few changes I'd request.

Don't include:

• scripts/groq_chat.exs
• scripts/groq_chat.sh

While cool, it's not something any of the other services are doing either.

In test/chat_models/chat_groq_test.exs, there's a test that does this:

      assert get_in(data, [:messages, Access.at(0), "role"]) == :user
      assert get_in(data, [:messages, Access.at(0), "content"]) == user_message
      assert get_in(data, [:messages, Access.at(1), "role"]) == :assistant
      assert get_in(data, [:messages, Access.at(1), "content"]) == assistant_message

I much prefer using pattern matches to assert a data shape because it's easier to look at a test and know what it's doing and testing.

Thanks!