Error using gradient_checkpoining

I am unable to train with gradient_checkpoining on
`RuntimeError: element 0 of variables does not require grad and does not have a grad_fn`
Any idea how to fix that?

Full training command:
```
accelerate launch --num_cpu_threads_per_process=2 "./train_network.py" --enable_bucket
--min_bucket_reso=256 --max_bucket_reso=1024
--pretrained_model_name_or_path="B:/AIimages/stable-diffusion-webui/models/Stable-diffusion/model.ckpt" --train_data_dir="B:\AIimages\training\data"
--resolution="576,576" --output_dir="B:\AIimages\stable-diffusion-webui\models\Lora\lora\out"
--logging_dir="B:\AIimages\stable-diffusion-webui\models\Lora\lora\out" --network_alpha="128"
--save_model_as=safetensors --network_module=networks.lora --unet_lr=0.0001
--network_train_unet_only --network_dim=128 --output_name="test_v1"
--lr_scheduler_num_cycles="48" --no_half_vae --learning_rate="1.0" --lr_scheduler="cosine"
--lr_warmup_steps="34" --train_batch_size="8" --max_train_steps="684" --save_every_n_epochs="1"
--mixed_precision="fp16" --save_precision="fp16" --seed="420" --caption_extension=".txt"
--cache_latents --cache_latents_to_disk --optimizer_type="AdamW8bit" --optimizer_args
betas=0.9,0.999 weight_decay=0.01 --max_data_loader_n_workers="0" --max_token_length=225
--clip_skip=2 --bucket_reso_steps=1 --min_snr_gamma=10 --flip_aug --shuffle_caption
--gradient_checkpointing --xformers --bucket_no_upscale --noise_offset=0.0
--sample_sampler=euler_a
--sample_prompts="B:\AIimages\stable-diffusion-webui\models\Lora\lora\out\sample\prompt.txt"
--sample_every_n_epochs="1"

```

Full error:

```
epoch 1/7
B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\torch\utils\checkpoint.py:31: UserWarning: None of the inputs have requires_grad=True. Gradients will be None
warnings.warn("None of the inputs have requires_grad=True. Gradients will be None")
╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ B:\AIimages\sd-scripts\kohya_ss\train_network.py:990 in │
│ │
│ 987 │ args = train_util.read_config_from_file(args, parser) │
│ 988 │ │
│ 989 │ trainer = NetworkTrainer() │
│ ❱ 990 │ trainer.train(args) │
│ 991 │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\train_network.py:798 in train │
│ │
│ 795 │ │ │ │ │ │
│ 796 │ │ │ │ │ loss = loss.mean() # 平均なのでbatch_sizeで割る必要なし │
│ 797 │ │ │ │ │ │
│ ❱ 798 │ │ │ │ │ accelerator.backward(loss) │
│ 799 │ │ │ │ │ if accelerator.sync_gradients and args.max_grad_norm != 0.0: │
│ 800 │ │ │ │ │ │ params_to_clip = network.get_trainable_params() │
│ 801 │ │ │ │ │ │ accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm) │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\accelerate\accelerator.py:1743 in │
│ backward │
│ │
│ 1740 │ │ elif self.distributed_type == DistributedType.MEGATRON_LM: │
│ 1741 │ │ │ return │
│ 1742 │ │ elif self.scaler is not None: │
│ ❱ 1743 │ │ │ self.scaler.scale(loss).backward(**kwargs) │
│ 1744 │ │ else: │
│ 1745 │ │ │ loss.backward(**kwargs) │
│ 1746 │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\torch\_tensor.py:487 in backward │
│ │
│ 484 │ │ │ │ create_graph=create_graph, │
│ 485 │ │ │ │ inputs=inputs, │
│ 486 │ │ │ ) │
│ ❱ 487 │ │ torch.autograd.backward( │
│ 488 │ │ │ self, gradient, retain_graph, create_graph, inputs=inputs │
│ 489 │ │ ) │
│ 490 │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\torch\autograd\__init__.py:200 in │
│ backward │
│ │
│ 197 │ # The reason we repeat same the comment below is that │
│ 198 │ # some Python versions print out the first line of a multi-line function │
│ 199 │ # calls in the traceback and some print out the last line │
│ ❱ 200 │ Variable._execution_engine.run_backward( # Calls into the C++ engine to run the bac │
│ 201 │ │ tensors, grad_tensors_, retain_graph, create_graph, inputs, │
│ 202 │ │ allow_unreachable=True, accumulate_grad=True) # Calls into the C++ engine to ru │
│ 203 │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn
steps: 0%| | 0/684 [00:01:7 │
│ │
│ 4 from accelerate.commands.accelerate_cli import main │
│ 5 if __name__ == '__main__': │
│ 6 │ sys.argv[0] = re.sub(r'(-script\.pyw|\.exe)?$', '', sys.argv[0]) │
│ ❱ 7 │ sys.exit(main()) │
│ 8 │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\accelerate\commands\accelerate_cli.py:45 │
│ in main │
│ │
│ 42 │ │ exit(1) │
│ 43 │ │
│ 44 │ # Run │
│ ❱ 45 │ args.func(args) │
│ 46 │
│ 47 │
│ 48 if __name__ == "__main__": │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py:918 in │
│ launch_command │
│ │
│ 915 │ elif defaults is not None and defaults.compute_environment == ComputeEnvironment.AMA │
│ 916 │ │ sagemaker_launcher(defaults, args) │
│ 917 │ else: │
│ ❱ 918 │ │ simple_launcher(args) │
│ 919 │
│ 920 │
│ 921 def main(): │
│ │
│ B:\AIimages\sd-scripts\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py:580 in │
│ simple_launcher │
│ │
│ 577 │ process.wait() │
│ 578 │ if process.returncode != 0: │
│ 579 │ │ if not args.quiet: │
│ ❱ 580 │ │ │ raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd) │
│ 581 │ │ else: │
│ 582 │ │ │ sys.exit(1) │
│ 583 │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
```

Start at train_network.py around the accelerator.backward(loss) call and inspect how the --gradient_checkpointing option reaches that training path. Reproduce with the supplied accelerate launch command, paying attention to the PyTorch warning that no inputs require gradients. Done means training proceeds past the first step without the reported RuntimeError.

Written by the indexing model from the issue text.