Questions related to the finetuning process of Flaubert on a dataset ?

[getalp]

Questions related to the finetuning process of Flaubert on a dataset ?

[formiel]

Hi @keloemma,

Question 1, 2, and 3: The goal of the script extract_split_cls.py is to prepare data in the proper format for the fine-tuning script. In particular, it extracts the text, get the labels, and split the training set into train and dev sets since there is no dev set provided separately for this task. So you should modify or create your own script to process your data if your data is different from the CLS dataset.

The format of the output files depend on the task. If you use Hugging Face's transformer library for finetuning, you can check out the details of the output format here if your task is similar to the ones in the GLUE benchmark and prepare your data accordingly.

There is also a "parse.py" script , At which level is it used in the flow of the architecture ?

It is included in the downloaded data but we do not use it in our code.

About the finetuning, what is the utility of the config file ? How is it obtained ?

The configuration file is used to save different training parameters. You can run your experiments with different configurations using the same running command. For more details, you can check out the configuration files in the examples and see the parameters. These parameters are used in the running command.

[schwabdidier]

I assume that you got your answer @keloemma ?