torch & accelerate version

[Hope7Happiness]

Hi, thank you for your great work!

I was running the code with torch 2.0.0 and accelerate 0.12.0. I have encountered two following errors:

1. When running bash_scripts/run.sh with the first line:

accelerate launch --num_processes 1 train_flow_latent.py --exp celeb256_f8_adm \
     --dataset celeba_256 --datadir ../cnf_flow/data/celeba/celeba-lmdb \
     --batch_size 112 --num_epoch 500 \
     --image_size 256 --f 8 --num_in_channels 4 --num_out_channels 4 \
     --nf 256 --ch_mult 1 2 2 2 --attn_resolution 16 8 --num_res_blocks 2 \
     --lr 2e-5 --scale_factor 0.18215 \
     --save_content --save_content_every 10 \
     --use_origin_adm

it seems that the --f 8 argument make it ambiguous, giving an error like this:

accelerate <command> [<args>] launch: error: ambiguous option: --f could match --fsdp_offload_params, --fsdp_min_num_params, --fsdp_sharding_strategy, --fsdp_auto_wrap_policy, --fsdp_transformer_layer_cls_to_wrap, --fsdp_backward_prefetch_policy, --fsdp_state_dict_type, --fp16

2. When adding an option --use_ema, there is another error:

Traceback (most recent call last):                                                         
  File "/srv/username/LFM/train_flow_latent.py", line 342, in <module>                     
    train(args)                                                                            
  File "/srv/username/LFM/train_flow_latent.py", line 93, in train                         
    data_loader, model, optimizer, scheduler = accelerator.prepare(data_loader, model, opti
mizer, scheduler)                                                                          
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/accelerate/accel
erator.py", line 621, in prepare                                                           
    result = tuple(self._prepare_one(obj, first_pass=True) for obj in args)
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/accelerate/accelerator.py", line 621, in <genexpr>
    result = tuple(self._prepare_one(obj, first_pass=True) for obj in args)
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/accelerate/accelerator.py", line 520, in _prepare_one
    optimizer = self.prepare_optimizer(obj)
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/accelerate/accelerator.py", line 854, in prepare_optimizer
    optimizer = AcceleratedOptimizer(optimizer, device_placement=self.device_placement, scaler=self.scaler)
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/accelerate/optim
izer.py", line 70, in __init__
    self.optimizer.load_state_dict(state_dict)
  File "/srv/username/LFM/EMA.py", line 66, in load_state_dict
    super(EMA, self).load_state_dict(state_dict)
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/torch/optim/opti
mizer.py", line 433, in load_state_dict
    self.__setstate__({'state': state, 'param_groups': param_groups})
  File "/srv/username/miniconda3/envs/LFMorig/lib/python3.10/site-packages/torch/optim/opti
mizer.py", line 214, in __setstate__
    self.defaults.setdefault('differentiable', False)
AttributeError: 'EMA' object has no attribute 'defaults'

Could you please kindly share your torch and accelerate version for running the scripts? Thank you!