ValueError: Query/Key/Value should all have the same dtype

[y0himba]

Not sure if this is a bug or some interaction with an extension and settings, so posting here to ask for advice. I am using the Animator tab which is created by the excellent and aptly named Animator Extension

When trying to create an animation, I get the following error.

ValueError: Query/Key/Value should all have the same dtype
  query.dtype: torch.float32
  key.dtype  : torch.float32
  value.dtype: torch.float16

Any help is appreciated.

[y0himba]

I figured this out. Settings-->Stable Diffusion.

[vlx]

I noticed it works after I changed the SD checkpoint in use (.safetensors file).

[blackken200]

thank you for this , it fixed!

[yqtop]

thank you so much

[Emefaitch]

i just encountered this problem after updating, was driving me mad, unchecked it and everything works again. Thank you so much for saving my sanity.

[khayamgondal]

I am using accelerate command line not automatic111. How do I pass the flag there?
@y0himba can you share your accelerate command with the 'upcast cross attention to 32' flag unchecked?

My current accelerate command

accelerate launch diffusers/examples/dreambooth/train_dreambooth_lora_sdxl.py \
      --pretrained_model_name_or_path="stabilityai/stable-diffusion-xl-base-1.0" \
      --instance_data_dir={input_dir} \
      --output_dir={output_dir} \
      --instance_prompt=instance_prompt \
      --mixed_precision="fp16" \
      --resolution=1024 \
      --train_batch_size=1 \
      --gradient_accumulation_steps=4 \
      --learning_rate=1e-4 \
      --lr_scheduler="constant" \
      --lr_warmup_steps=0 \
      --checkpointing_steps=500 \
      --max_train_steps=1000 \
      --seed="0" \
      --checkpoints_total_limit=5 \
      --enable_xformers_memory_efficient_attention