AssertionError: You can't use same Accelerator() instance with multiple models when using DeepSpeed

[dizhenx]

Describe the bug

The same configuration, training on an 8 GB GPU, can work fine, but when executing the script to fine-tune the text encoder with UNet, an error occurs.
The config file is as follows:
compute_environment: LOCAL_MACHINE
deepspeed_config:
gradient_accumulation_steps: 1
gradient_clipping: 1.0
offload_optimizer_device: cpu
offload_param_device: cpu
zero3_init_flag: true
zero_stage: 2
distributed_type: DEEPSPEED
downcast_bf16: 'no'
dynamo_config:
dynamo_backend: FX2TRT
machine_rank: 0
main_training_function: main
mixed_precision: fp16
num_machines: 1
num_processes: 4
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
The error message is as follows:
You can't use same Accelerator() instance with multiple models when using DeepSpeed

Reproduction

accelerate launch --config_file /mnt/data/huggingface/accelerate/default_config1.yaml train_dreambooth.py --pretrained_model_name_or_path=$MODEL_NAME --train_text_encoder --instance_data_dir=$INSTANCE_DIR --class_data_dir=$CLASS_DIR --output_dir=$OUTPUT_DIR --with_prior_preservation --prior_loss_weight=1.0 --instance_prompt="a photo of sks dog" --class_prompt="a photo of dog" --resolution=512 --train_batch_size=1 --use_8bit_adam --gradient_checkpointing --learning_rate=2e-6 --lr_scheduler="constant" --lr_warmup_steps=0 --num_class_images=200 --max_train_steps=20

Logs

╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ /mnt/data/creative/diffusers/examples/dreambooth/train_dreambooth.py:869 in <module>             │
│                                                                                                  │
│   866                                                                                            │
│   867 if __name__ == "__main__":                                                                 │
│   868 │   args = parse_args()                                                                    │
│ ❱ 869 │   main(args)                                                                             │
│   870                                                                                            │
│                                                                                                  │
│ /mnt/data/creative/diffusers/examples/dreambooth/train_dreambooth.py:684 in main                 │
│                                                                                                  │
│   681 │   )                                                                                      │
│   682 │                                                                                          │
│   683 │   if args.train_text_encoder:                                                            │
│ ❱ 684 │   │   unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prep   │
│   685 │   │   │   unet, text_encoder, optimizer, train_dataloader, lr_scheduler                  │
│   686 │   │   )                                                                                  │
│   687 │   else:                                                                                  │
│                                                                                                  │
│ /mnt/data/creative/miniconda3/envs/diffusers/lib/python3.9/site-packages/accelerate/accelerator. │
│ py:1148 in prepare                                                                               │
│                                                                                                  │
│   1145 │   │   │   │   if isinstance(obj, torch.nn.Module):                                      │
│   1146 │   │   │   │   │   model_count += 1                                                      │
│   1147 │   │   │   if model_count > 1:                                                           │
│ ❱ 1148 │   │   │   │   raise AssertionError(                                                     │
│   1149 │   │   │   │   │   "You can't use same `Accelerator()` instance with multiple models wh  │
│   1150 │   │   │   │   )                                                                         │
│   1151                                                                                           │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
AssertionError: You can't use same `Accelerator()` instance with multiple models when using DeepSpeed
[10:57:19] ERROR    failed (exitcode: 1) local_rank: 0 (pid: 1146) of binary: /mnt/data/creative/miniconda3/envs/diffusers/bin/python3.9

System Info

• diffusers version: 0.15.0.dev0
• Platform: Linux-3.10.0-1160.92.1.el7.x86_64-x86_64-with-glibc2.17
• Python version: 3.9.2
• PyTorch version (GPU?): 2.0.1+cu117 (True)
• Huggingface_hub version: 0.16.2
• Transformers version: 4.27.1
• Accelerate version: 0.20.3
• xFormers version: not installed
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: yes