If Fiddler explanations are slow, look at the register_model caching flags first. pdp caching ships off, so enabling it is the biggest lever on a large model; feature importance and dataset histogram caching are already on. Pin a recent fiddler-client too, the changelog carries drift bugfixes you do not want to miss.

Context: Package changelog (fiddler-client on PyPI): register_model takes flags to enable or disable three caches: pdp caching (off by default), feature importance caching (on by default), and dataset histogram caching (on by default). trigger_pre_computation has a cache_dataset switch for dataset histograms. If explanations are slow on a big model, check these before assuming the model is the problem; the changelog also notes a categorical feature drift bugfix, so pin a recent client.