UNet cannot be saved in `diffusers` nor `diffusers_safetensors` format

Hi,
I was running the script on Google Colab via the following command:
```bash
!accelerate launch --num_cpu_threads_per_process 1 sd-scripts/train_db.py \
--pretrained_model_name_or_path=$MODEL_NAME \
--dataset_config=$repo_path_conc/initial.toml \
--output_dir=$OUTPUT_DIR \
--save_model_as=diffusers \
--prior_loss_weight=1.0 \
--max_train_steps=$SAVE_MAX_STEPS \
--save_every_n_steps=$SAVE_INTERVAL \
--learning_rate=3.4e-7 \
--lr_scheduler="cosine" \
--optimizer_type="AdamW8bit" \
--xformers \
--mixed_precision="fp16" \
--gradient_checkpointing \
--sample_every_n_steps=$SAVE_INTERVAL \
--sample_sampler="k_euler" \
--sample_prompts="/content/prompts-initial.txt"
```
Everything went fine, the samples became more and more pretty, the loss found its minimum... but at the end of training I noticed absence of the main part - UNet files!


What am I doing wrong?

Start with sd-scripts/train_db.py and trace how --save_model_as=diffusers and diffusers_safetensors are handled during the save path. Reproduce the Google Colab command or a reduced training run, then verify that the resulting output includes the UNet files in the requested format.

Written by the indexing model from the issue text.