I noticed that language model data does not auto-suggest vocab when you tab data_lm. with v (not even in suggested list), but works fine if typed manually. Is this just me or a small bug?
The other magic I hope to see is **kwargs expansion when you shift-tab - how hard is this?
It’s not auto-completed because it’s actually an attribute of data_lm.train_ds, but when it doesn’t find an attibute, data_lm automatically delegates to its train_ds.
Thanks @sgugger for clarifying. Now I see it in docs (was looking at v3 notebooks before). What about kwargs expansion? I know they are checked at some stage but would help to view them ahead.
Ex. https://docs.fast.ai/basic_train.html#fit_one_cycle there is ** kwargs but how would one know what to expect if not listed or linked?
For now the only way to see what those kwargs can be is to follow where they are passed in the source code. In this case, they are given to OneCycleScheduler.