The fixes in this module monkeypatched `torch` to resolve some issues with FP16 on macOS. These issues have long since been resolved.
Included in the now-removed fixes is `CustomSlicedAttentionProcessor`, which is intended to reduce memory requirements for MPS. This overrides `diffusers`' own `SlicedAttentionProcessor`.
Unfortunately, `attention_type: sliced` produces hot garbage with the fixes and black images without the fixes. So this class appears to now be a moot point.
Regardless, SDPA is supported on MPS and very efficient, so sliced attention is largely obsolete.
In `ObjectSerializerDisk`, we use `torch.load` to load serialized objects from disk. With torch 2.6.0, torch defaults to `weights_only=True`. As a result, torch will raise when attempting to deserialize anything with an unrecognized class.
For example, our `ConditioningFieldData` class is untrusted. When we load conditioning from disk, we will get a runtime error.
Torch provides a method to add trusted classes to an allowlist. This change adds an arg to `ObjectSerializerDisk` to add a list of safe globals to the allowlist and uses it for both `ObjectSerializerDisk` instances.
Note: My first attempt inferred the class from the generic type arg that `ObjectSerializerDisk` accepts, and added that to the allowlist. Unfortunately, this doesn't work.
For example, `ConditioningFieldData` has a `conditionings` attribute that may be one some other untrusted classes representing model-specific conditioning data. So, even if we allowlist `ConditioningFieldData`, loading will fail when torch deserializes the `conditionings` attribute.
Before FLUX Fill was merged, we didn't do any checks for the model variant. We always returned "normal".
To determine if a model is a FLUX Fill model, we need to check the state dict for a specific key. Initially, this logic was too strict and rejected quantized FLUX models. This issue was resolved, but it turns out there is another failure mode - some fine-tunes use a different key.
This change further reduces the strictness, handling the alternate key and also falling back to "normal" if we don't see either key. This effectively restores the previous probing behaviour for all FLUX models.
Closes#7856Closes#7859
In #7780 we added FLUX Fill support, and needed the probe to be able to distinguish between "normal" FLUX models and FLUX Fill models.
Logic was added to the probe to check a particular state dict key (input channels), which should be 384 for FLUX Fill and 64 for other FLUX models.
The new logic was stricter and instead of falling back on the "normal" variant, it raised when an unexpected value for input channels was detected.
This caused failures to probe for BNB-NF4 quantized FLUX Dev/Schnell, which apparently only have 1 input channel.
After checking a variety of FLUX models, I loosened the strictness of the variant probing logic to only special-case the new FLUX Fill model, and otherwise fall back to returning the "normal" variant. This better matches the old behaviour and fixes the import errors.
Closes#7822
