I am not very familiar with the code, but default mutable values for function arguments (like includes=[] here) can cause problems. https://docs.quantifiedcode.com/python-anti-patterns/correctness/mutable_default_value_as_argument.html

should includes be a dictionary or a list? It was originally a list, but here you're setting it to a dict if not set. It looks like when you use it you're treating it as a list.
You can use [] here instead of list() too (or {} instead of dict())

Sorry, my bad. Fixed.

This is a little complex to understand - I think that you're passing in includes_data as an argument and it's getting populated in get_relationship_links. Is that correct?
If that is the case, perhaps we shouldn't call the function get_...... The documentation should be updated to say that this value is altered, instead of being passed in as an argument.

Yes, includes_data is getting populated in get_relationship_links. I have changed the name of the function to _relationship_links_helper (the function would not be used outside this module). Also, as suggested, I added a Note in the docstring stating that includes_data is altered to contain the relationship objects keyed by the mbid.

Sometimes it's useful to use collections.defaultdict for things like this, where you want to be able to add arbitrary nested objects. Note that this might not serialise nicely in your redis cache and so you might need to convert it before you cache it.

Thanks. Updated as suggested.

It might be a good idea to make these arguments keyword-only to make it more explicit when this function is used.

Would a name like get_relationship_info be more informative for this function?

Also if this function doesn't return anything, what does it do? Update one of the arguments passed to it? Can you describe this (in the docstring please)?

This is not really a standard way of writing an extended description. Check https://github.com/metabrainz/guidelines/blob/master/Python.md#docstrings. Here's how you could write this:

    """Get relationship links between two entities.

    Keep in mind that includes_data (dict) is altered to contain the relationship objects
    keyed by the source entity MBIDs.

    Args:
        relation (mbdata.model): Model relating the two entities.
        query (Session.query): Query object.
        source_attr (str): 'entity0' or 'entity1' based on which represents source model in relation table.
        target_attr (str): 'entity0' or 'entity1' based on which represents target model in relation table.
        target_type (str): Type of the target entity.
        source_entity_ids (list): IDs of the source entity.
        includes_data (dict): Dictionary containing the includes data of entities.
    """

In this case mbids can actually be made not keyword-only. It's clear how fetch_multiple_places([<mbid_1>, <mbid_2>]) would work.