Add Pillow >= 9.4 requirement

[ebsmothers]

torchvision nightlies come with PIL == 9.3 (at least in my env), but unless we have PIL >= 9.4 we will run into errors on our multimodal datasets around the usage of ExIfTags (ref). Ideally we can change this in torchvision, but until then gonna pin to a minimum version of PIL just to be safe.

Test plan:

conda create -n test-pillow python=3.11 -y
conda activate test-pillow
pip install --pre torch torchvision torchao --index-url https://download.pytorch.org/whl/nightly/cu121
pip list
...
Pillow                   9.3.0
...

# From torchtune directory
pip install -e .
pip list
...
pillow                   10.4.0
...

Before:

  File "/home/ebs/.conda/envs/nightly-testing-09-25/lib/python3.11/site-packages/PIL/Image.py", line 76, in __getattr__
    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")
AttributeError: module 'PIL.Image' has no attribute 'ExifTags'

After:

1|1|Loss: 0.2959183156490326:   0%|▍                                                                                                                           | 1/304 [00:27<2:20:33, 27.83s/it]

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/1671

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit fe9a73f with merge base 55f29f7 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[RdoubleA]

Can you confirm that the original error doesn't happen anymore on 9.4?