In dask_ml linear_models LinearRegression. If you use model.score(X_test, y_test) then you get back the Mean Squared Error instead of the R2 which is how sklearn's implementation does it.
On top of that the score method defined has a docstring that states that it returns the r2 score instead of the MSE.

We can see in the code here

That the docstring says r2 score, but the method returns the mean_squared_error from dask_ml.utils.

Either the docstring should be changed to reflect the real return value. Or the r2_score method should be defined and then use that on the rerturn statement of the score method