Allow "auto" for gradient clipping in YAML

[regisss]

What does this PR do?

If we have gradient_clipping: auto in the DeepSpeed config inside an Accelerate config, the code currently fails with:

Traceback (most recent call last):
  File "/root/workspace/lora/tmp.py", line 3, in <module>
    args = TrainingArguments(
  File "<string>", line 124, in __init__
  File "/usr/local/lib/python3.10/dist-packages/transformers/training_args.py", line 1828, in __post_init__
    self.deepspeed_plugin = DeepSpeedPlugin()
  File "<string>", line 14, in __init__
  File "/usr/local/lib/python3.10/dist-packages/accelerate/utils/dataclasses.py", line 727, in __post_init__
    self.gradient_clipping = float(gradient_clipping)
ValueError: could not convert string to float: 'auto'

This can be reproduced running accelerate launch --config_file my_config.yaml script.py with script.py being

from transformers import TrainingArguments, Trainer, AutoModel

args = TrainingArguments(
    output_dir="/tmp/test",
    max_grad_norm=0.5,
)

trainer = Trainer(
    model=AutoModel.from_pretrained("bert-base-uncased"),
    args=args,
)

and my_config.yaml is

compute_environment: LOCAL_MACHINE
debug: false
deepspeed_config:
  gradient_accumulation_steps: 1
  gradient_clipping: auto
  offload_optimizer_device: none
  offload_param_device: none
  zero3_init_flag: true
  zero3_save_16bit_model: true
  zero_stage: 3
distributed_type: DEEPSPEED
downcast_bf16: 'no'
machine_rank: 0
main_training_function: main
mixed_precision: bf16
num_machines: 1
num_processes: 8
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false

This PR fixes this issue by checking if the given value can actually be casted to float.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[muellerzr]

cc @pacman100

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[pacman100]

Thank you @regisss for supporting auto for gradient_clipping to be inline with gradient_accumulation_steps providing more flexibility to users! Left a suggestion.

[pacman100]

Thank you @regisss!