[utc] fix loading local model in taskflow

[LemonNoel]

PR types

Bug fixes

PR changes

APIs

Description

• Fix bug when load local checkpoints from task_path in taskflow.

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Merging #4505 (7f51676) into develop (689428a) will increase coverage by 0.01%.
The diff coverage is 79.41%.

@@             Coverage Diff             @@
##           develop    #4505      +/-   ##
===========================================
+ Coverage    41.27%   41.29%   +0.01%     
===========================================
  Files          432      432              
  Lines        61705    61733      +28     
===========================================
+ Hits         25468    25491      +23     
- Misses       36237    36242       +5     

Impacted Files	Coverage Δ
...addlenlp/taskflow/zero_shot_text_classification.py	18.25% <33.33%> (+0.79%)	⬆️
paddlenlp/utils/serialization.py	88.05% <75.00%> (-0.23%)	⬇️
paddlenlp/transformers/t5/modeling.py	84.82% <86.95%> (-0.26%)	⬇️
paddlenlp/transformers/nezha/modeling.py	20.60% <0.00%> (-0.34%)	⬇️
paddlenlp/transformers/activations.py	78.68% <0.00%> (+1.63%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[linjieccc]

@LemonNoel 改成from_pretrained("{local_path}")的形式后，需要定义resource_files_names和resource_files_urls并在__init__ 中增加 self._check_task_files()，可以参考这里https://github.com/PaddlePaddle/PaddleNLP/blob/develop/paddlenlp/taskflow/information_extraction.py#L115

关于例如from_pretrained("utc_large")调用后模型不能再通过from_pretrained("{local_path}")方式加载的问题也请 @wj-Mcat 帮忙看下，我们后续看看能不能解决一下这里的gap

[sijunhe]

如果是改成这样，没必要if了，直接一行model_instance = UTC.from_pretrained(self._task_path, from_hf_hub=self.from_hf_hub) 就行了。
@linjieccc UTC taskflow有做from_pretrained以外的文件下载吗？像这种已经整合pretrained config的模型和taskflow, 建议下载功能全部由from_pretrained承载，不要再做分开的下载逻辑了

[linjieccc]

@sijunhe 嗯嗯，这里确实改成下载功能全部通过from_pretrained承载好些，也可以避免模型文件重复下载的情况，后续会针对这块处理进行统一升级

关于Taskflow内非transformers类的模型，例如像GRU-CRF，目前模型是放在$PPNLP_HOME/.taskflow/{task_name}/{model_name}，是否这部分模型后续也统一放在$PPNLP_HOME/.paddlenlp/models管理，模型加载的代码在Taskflow中实现

[sijunhe]

任何涉及LEGACY_CONFIG_NAME的都可以删除，因为UTC在ernie升级pretrainded config之后，不涉及向后兼容 @LemonNoel

[LemonNoel]

任何涉及LEGACY_CONFIG_NAME的都可以删除，因为UTC在ernie升级pretrainded config之后，不涉及向后兼容 @LemonNoel

已删除

[linjieccc]

LGTM