Add RELLM experimental LLM decoding

[vowelparrot]

RELLM is a library that wraps local HuggingFace pipeline models for structured decoding.

It works by generating tokens one at a time. At each step, it masks tokens that don't conform to the provided partial regular expression.

Checked it out and got a small HuggingFace model to generate structured format easily :)

[eyurtsev]

@vowelparrot Is the output from the first cell correct? I think it's supposed to illustrate that the response that's generated isn't correct, but I don't see it generating an auto-completion response

[eyurtsev]

Looks good to me @vowelparrot, @nfcampos let a comment.

[vowelparrot]

@vowelparrot Is the output from the first cell correct? I think it's supposed to illustrate that the response that's generated isn't correct, but I don't see it generating an auto-completion response

Ya the model just repeats the question there. It's a really small model to fit on my laptop