[docs] Quantization + torch.compile + offloading

[stevhliu]

Follows up on #11670 and #11672 to document combinations of quantization, torch.compile, and offloading.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[stevhliu]

@sayakpaul, can you help with the numbers here?

[sayakpaul]

Thanks for starting this. Will get you the numbers.

[sayakpaul]

@stevhliu

combination	latency	memory usage
quantization	32.602	14.9453
quantization, torch.compile	25.847	14.9448
quantization, torch.compile, model CPU offloading	32.312	12.2369
quantization, torch.compile, group offloading	60.235	12.2369

Code: https://gist.github.com/sayakpaul/0db9d8eeeb3d2a0e5ed7cf0d9ca19b7d

Worth mentioning:

• We are applying quantization to transformer and text_encoder_2
• GPU used: RTX 4090
• Using PyTorch nightlies is better
• https://huggingface.co/docs/diffusers/main/en/optimization/memory#group-offloading mentions why the speed-memory trade-off with group offloading in Flux isn't as expected.