Rémi Louf

This should be kept out of the model class, which should only check the validity of the data it is fed.

Try the Ornstein-Uhlenbeck kernel for the gaussian processes

3

The Ornstein-Uhlenbeck kernel defines a stationary gaussian process that describes a mean-reverting stochastic process. It seems more adapted to our modeling problem than a gaussian kernel that doesn't have an...

Voter uncertainty can change the conclusions dramatically, especially early in the campaign where close to 50% of poll respondants say they could still change their mind. It is thus important...

Which would allow us to study how intentions flow between candidates from one poll to the next. This is especially important in an election with 2 rounds.

Test the CFG integration for `llama.cpp`

3

I took a quick look at the `llama.cpp` CFG integration and it doesn't seem to give the correct results. We need to investigate and understand what is wrong.

Use vLLM's native Outlines integration for `outlines.serve.serve`

4

Now that Outlines was integrated in vLLM we can remove the server code from Outlines and instead import vLLM's code for `outlines.serve.serve`, which should simplify the codebase and ensure that...

[Pydantic does this](https://github.com/pydantic/pydantic/blob/main/.github/workflows/ci.yml). Not only would it save time, it would also be incentive to be more verbose in release notes.

Add Outlines in LangChain

5

We should open a PR on the LangChain repo to add Outlined as a model / guided generation provider.