huggingface/transformers#12276 introduced a new --log_level feature, which now allows users to set their desired log level via CLI or TrainingArguments.

run_translation.py was used as a "model" for other examples.

Now we need to replicate this to all other Trainer-based examples under examples/pytorch/, the 3 changes are

1. importing datasets
2. using `training

建议在readme里面添加预训练模型环境变量地址修改的提示。
我之前遇到过磁盘空间不够，原因就是这个问题
但不是所有人都会debug源码去查看框架读的是哪个系统变量

The code would already produce charts in English. However, the current charts are in French.

One would need to try all the exercises to regenerate new charts, incorporate them in the notebook, and then regenerate the README in a clean way, as well as the .py file if the code in the notebook changed.

The French branch (which is named francais) should remain untouched.