'Trainer' object has no attribute '_accelerator_connector' when loaded from a pickle

[CompRhys]

I use torch.save to pickle a sklearn Pipeline containing a pl.Trainer object. When I try to use this Trainer as part of my pipeline it says that it does not have a '_accelerator_connector'.

This seems surprising to me and I haven't found any other posts about this issue.

Input In [19], in <cell line: 1>()
----> 1 form_pipe["regression_model"].models["target"].trainer.strategy

File ~/opt/miniconda3/envs/base/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py:2007, in Trainer.strategy(self)
   [2005](file:///Users/comprhys/opt/miniconda3/envs/base/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py?line=2004) @property
   [2006](file:///Users/comprhys/opt/miniconda3/envs/base/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py?line=2005) def strategy(self) -> Strategy:
-> [2007](file:///Users/comprhys/opt/miniconda3/envs/baselib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py?line=2006)     return self._accelerator_connector.strategy

AttributeError: 'Trainer' object has no attribute '_accelerator_connector'

[gioannides]

any follow up on this?

[CompRhys]

In the end I ended up just pickling the nested model rather torch using torch save -- not ideal but it worked

[gioannides]

I pickled mine and it still doesn't work.