[docs] fix bug in deepspeed config

[faaany]

What does this PR do?

While running the deepspeed example code, I got the following error, indicating the value of stage3_prefetch_bucket_size in the deepspeed config is not correct. The given value is 3774873.6, which is a float, but it should be an integer. So this PR fixes this.

[rank0]: Traceback (most recent call last):
[rank0]:   File "/home/sdp/fanli/doc_to_fix.py", line 112, in <module>
[rank0]:     model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
[rank0]:   File "/home/sdp/fanli/transformers/src/transformers/models/auto/auto_factory.py", line 564, in from_pretrained
[rank0]:     return model_class.from_pretrained(
[rank0]:   File "/home/sdp/fanli/transformers/src/transformers/modeling_utils.py", line 261, in _wrapper
[rank0]:     return func(*args, **kwargs)
[rank0]:   File "/home/sdp/fanli/transformers/src/transformers/modeling_utils.py", line 4144, in from_pretrained
[rank0]:     deepspeed.zero.Init(config_dict_or_path=deepspeed_config()),
[rank0]:   File "/home/sdp/miniforge3/envs/ipex-ww04/lib/python3.10/site-packages/deepspeed-0.16.3+66d3d3e94-py3.10.egg/deepspeed/runtime/zero/partition_parameters.py", line 949, in __init__
[rank0]:     _ds_config = deepspeed.runtime.config.DeepSpeedConfig(config_dict_or_path,
[rank0]:   File "/home/sdp/miniforge3/envs/ipex-ww04/lib/python3.10/site-packages/deepspeed-0.16.3+66d3d3e94-py3.10.egg/deepspeed/runtime/config.py", line 797, in __init__
[rank0]:     self._initialize_params(copy.copy(self._param_dict))
[rank0]:   File "/home/sdp/miniforge3/envs/ipex-ww04/lib/python3.10/site-packages/deepspeed-0.16.3+66d3d3e94-py3.10.egg/deepspeed/runtime/config.py", line 817, in _initialize_params
[rank0]:     self.zero_config = get_zero_config(param_dict)
[rank0]:   File "/home/sdp/miniforge3/envs/ipex-ww04/lib/python3.10/site-packages/deepspeed-0.16.3+66d3d3e94-py3.10.egg/deepspeed/runtime/zero/config.py", line 73, in get_zero_config
[rank0]:     return DeepSpeedZeroConfig(**zero_config_dict)
[rank0]:   File "/home/sdp/miniforge3/envs/ipex-ww04/lib/python3.10/site-packages/deepspeed-0.16.3+66d3d3e94-py3.10.egg/deepspeed/runtime/config_utils.py", line 57, in __init__
[rank0]:     super().__init__(**data)
[rank0]:   File "/home/sdp/.local/lib/python3.10/site-packages/pydantic/main.py", line 214, in __init__
[rank0]:     validated_self = self.__pydantic_validator__.validate_python(data, self_instance=self)
[rank0]: pydantic_core._pydantic_core.ValidationError: 1 validation error for DeepSpeedZeroConfig
[rank0]: stage3_prefetch_bucket_size
[rank0]:   Input should be a valid integer, got a number with a fractional part [type=int_from_float, input_value=3774873.6, input_type=float]
[rank0]:     For further information visit https://errors.pydantic.dev/2.10/v/int_from_float

Documentation: @stevhliu

[stevhliu]

Thanks! Also pinging @muellerzr for a quick look :)

[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.

[faaany]

Hi @muellerz, could you pls take a look? Thx!