Fastai version 2.7.14 produces "AttributeError: device" when running .predict

[mumbledenoise]

Please confirm you have the latest versions of fastai, fastcore, and nbdev prior to reporting a bug (delete one): YES

Describe the bug
Running a prediction on Fastai 2.7.14 produces "AttributeError: device" when running .predict. Downgrading to 2.7.13 fixes it. I'm guessing this is because I'm using torch 1.13.1.

To Reproduce
Steps to reproduce the behavior:

Here are the logs of me reproducing the behavior, along with the fix:

root@Ubuntu-2204-jammy-amd64-base ~/run/gpu # python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from fastai.text.all import *
>>> sent = load_learner('model_3.pkl', cpu=False).to_fp16()
>>> sent.predict('hello')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.10/dist-packages/fastai/learner.py", line 320, in predict
    dl = self.dls.test_dl([item], rm_type_tfms=rm_type_tfms, num_workers=0)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/core.py", line 526, in test_dl
    return self.valid.new(test_ds, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/fastai/text/data.py", line 218, in new
    return super().new(dataset=dataset, res=res, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/core.py", line 94, in new
    res = super().new(dataset, cls, do_setup=False, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/load.py", line 144, in new
    bs=self.bs, shuffle=self.shuffle, drop_last=self.drop_last, indexed=self.indexed, device=self.device)
  File "/usr/local/lib/python3.10/dist-packages/fastcore/basics.py", line 496, in __getattr__
    if attr is not None: return getattr(attr,k)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/core.py", line 450, in __getattr__
    def __getattr__(self,k): return gather_attrs(self, k, 'tls')
  File "/usr/local/lib/python3.10/dist-packages/fastcore/transform.py", line 173, in gather_attrs
    if not res: raise AttributeError(k)
AttributeError: device

>>> exit()
root@Ubuntu-2204-jammy-amd64-base ~/run/gpu # pip show fastcore
Name: fastcore
Version: 1.5.29
Summary: Python supercharged for fastai development
Home-page: https://github.com/fastai/fastcore/
Author: Jeremy Howard and Sylvain Gugger
Author-email: infos@fast.ai
License: Apache Software License 2.0
Location: /usr/local/lib/python3.10/dist-packages
Requires: packaging, pip
Required-by: fastai, fastdownload, ohmeow-blurr

root@Ubuntu-2204-jammy-amd64-base ~/run/gpu # pip show fastai
Name: fastai
Version: 2.7.14
Summary: fastai simplifies training fast and accurate neural nets using modern best practices
Home-page: https://github.com/fastai/fastai
Author: Jeremy Howard, Sylvain Gugger, and contributors
Author-email: info@fast.ai
License: Apache Software License 2.0
Location: /usr/local/lib/python3.10/dist-packages
Requires: fastcore, fastdownload, fastprogress, matplotlib, packaging, pandas, pillow, pip, pyyaml, requests, scikit-learn, scipy, spacy, torch, torchvision
Required-by: fastinference, ohmeow-blurr

root@Ubuntu-2204-jammy-amd64-base ~/run/gpu # pip show torch
Name: torch
Version: 1.13.1
Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
Home-page: https://pytorch.org/
Author: PyTorch Team
Author-email: packages@pytorch.org
License: BSD-3
Location: /usr/local/lib/python3.10/dist-packages
Requires: nvidia-cublas-cu11, nvidia-cuda-nvrtc-cu11, nvidia-cuda-runtime-cu11, nvidia-cudnn-cu11, typing-extensions
Required-by: fastai, ohmeow-blurr, sentence-transformers, torchvision

root@Ubuntu-2204-jammy-amd64-base ~/run/gpu # pip install fastai==2.7.13
Collecting fastai==2.7.13
  Downloading fastai-2.7.13-py3-none-any.whl (232 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 232.2/232.2 KB 8.0 MB/s eta 0:00:00
Requirement already satisfied: pyyaml in /usr/lib/python3/dist-packages (from fastai==2.7.13) (5.4.1)
Requirement already satisfied: pillow>=9.0.0 in /usr/local/lib/python3.10/dist-packages (from fastai==2.7.13) (10.2.0)
Requirement already satisfied: scipy in /usr/local/lib/python3.10/dist-packages (from fastai==2.7.13) (1.12.0)
Requirement already satisfied: matplotlib in /usr/local/lib/python3.10/dist-packages (from fastai==2.7.13) (3.8.3)

...... (removed all of the Requirement already satisfied: lines for brevity)

Installing collected packages: fastai
  Attempting uninstall: fastai
    Found existing installation: fastai 2.7.14
    Uninstalling fastai-2.7.14:
      Successfully uninstalled fastai-2.7.14
Successfully installed fastai-2.7.13
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv

root@Ubuntu-2204-jammy-amd64-base ~/run/gpu # python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from fastai.text.all import *
>>> sent = load_learner('model_3.pkl', cpu=False).to_fp16()
>>> sent.predict('hello')
((0.01348114013671875,), tensor([0.0135]), tensor([0.0135]))
>>> 

Expected behavior
It either does the prediction, or it 2.7.14 marks torch 2.2 as being required (presumably that'd also have fixed it, since 2.7.14 brought torch 2.2 support, but I did not test this).

Error with full stack trace

Place between these lines with triple backticks:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.10/dist-packages/fastai/learner.py", line 320, in predict
    dl = self.dls.test_dl([item], rm_type_tfms=rm_type_tfms, num_workers=0)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/core.py", line 526, in test_dl
    return self.valid.new(test_ds, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/fastai/text/data.py", line 218, in new
    return super().new(dataset=dataset, res=res, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/core.py", line 94, in new
    res = super().new(dataset, cls, do_setup=False, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/load.py", line 144, in new
    bs=self.bs, shuffle=self.shuffle, drop_last=self.drop_last, indexed=self.indexed, device=self.device)
  File "/usr/local/lib/python3.10/dist-packages/fastcore/basics.py", line 496, in __getattr__
    if attr is not None: return getattr(attr,k)
  File "/usr/local/lib/python3.10/dist-packages/fastai/data/core.py", line 450, in __getattr__
    def __getattr__(self,k): return gather_attrs(self, k, 'tls')
  File "/usr/local/lib/python3.10/dist-packages/fastcore/transform.py", line 173, in gather_attrs
    if not res: raise AttributeError(k)
AttributeError: device

[steinmn]

Same behaviour seen on my setup with torch 2.2. Could the issue be related to the model being loaded having originally been generated with older library versions?