Bugfix: __name__ undefined on HuggingFace Pipeline Models.

[JosephCatrambone]

Traces are attempting to read from name and module, which aren't necessarily present on models. Given that this doesn't impact the control flow or anything outside of telemetry, it's safe to use sensible defaults for these values if they're not present, rather than throwing an exception.

Creating a draft PR as it would be wise to add additional tests to make sure we're handling things properly. Existing tests pass.

[JosephCatrambone]

I have some tests that I've added but haven't pushed. It has proved to be very challenging to authentically mock a pipeline/model/tokenizer, so (locally) I added a 96kb model + 100kb tokenizer to the mocks. We could pull from HF every time if we wanted to save the storage overhead, but given the nature of the system I think it's an okay addition to the tests. Are there any objections before I push the update to the PR?

[zsimjee]

I have some tests that I've added but haven't pushed. It has proved to be very challenging to authentically mock a pipeline/model/tokenizer, so (locally) I added a 96kb model + 100kb tokenizer to the mocks. We could pull from HF every time if we wanted to save the storage overhead, but given the nature of the system I think it's an okay addition to the tests. Are there any objections before I push the update to the PR?

Should be fine imo. Not huge.

[JosephCatrambone]

Added the tests and mock model.