Exp-001: Finetune gpt-2 model with new tokenizer on fr

[yongzx]

Added the following files to exp-001 folder.

• download_oscar_fr[.py, .sh]: for downloading oscar's fr training dataset.
• train_tokenizer_gpt2.py: for training GPT-2 tokenizer on the fr dataset.
• run_clm[.py, .sh]: for fine-tuning GPT-2

Let me know if I should change the folder name – I like to use numbering for experimental runs and then indicate in the README.md file about the descriptions of the folder.

Will include the training result (log file) and checkpoints once they are ready. #2

[yongzx]

Added

• run_clm_no_tok.sh: for finetuning GPT-2 with original tokenizer

@hadyelsahar I have requested access to the org https://huggingface.co/bigscience to use the bigscience/gpt2-350m-en checkpoint.