Describe the bug

decoding with enable_vae_tiling fails with the Stable Diffusion 2.1 model on certain sizes, throwing an IndexError.

Reproduction

import diffusers, torch
pipeline = diffusers.StableDiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16, revision="fp16").to("cuda")
pipeline.enable_xformers_memory_efficient_attention()
pipeline.enable_vae_tiling()

t0 = torch.zeros((1, 4, 20, 24))

with torch.inference_mode():
    decoded = pipeline.decode_latents(t)

Logs

IndexError: index 128 is out of bounds for dimension 2 with size 128

System Info

• diffusers version: 0.14.0
• Platform: Linux-5.15.0-67-generic-x86_64-with-glibc2.35
• Python version: 3.10.6
• PyTorch version (GPU?): 1.13.1+cu117 (True)
• Huggingface_hub version: 0.11.1
• Transformers version: 4.26.1
• Accelerate version: 0.16.0
• xFormers version: 0.0.16
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: no