Adding performer fine-tuning research exampke (#9239)

* added run_mlm_performer.py research example

* make styke

* make styke

* Added a README !

examples/research_projects/performer/README.md

+# Performer fine-tuning
+
+Example authors: @TevenLeScao, @Patrickvonplaten
+
+Paper authors: Krzysztof Choromanski, Valerii Likhosherstov, David Dohan, Xingyou Song, Andreea Gane, Tamas Sarlos, Peter Hawkins, Jared Davis, Afroz Mohiuddin, Lukasz Kaiser, David Belanger, Lucy Colwell, Adrian Weller
+
+## Requirements
+
+`datasets`, `flax` and `jax`. `wandb` integration is built-in if you want to use it.
+
+## Examples
+
+`sanity_script.sh` will launch performer fine-tuning from the bert-base-cased checkpoint on the Simple Wikipedia dataset (a small, easy-language English Wikipedia) from `datasets`.
+`full_script.sh` will launch performer fine-tuning from the bert-large-cased checkpoint on the English Wikipedia dataset from `datasets`.
+
+Here are a few key arguments:
+- Remove the `--performer` argument to use a standard Bert model.
+  
+- Add `--reinitialize` to start from a blank model rather than a Bert checkpoint. 
+  
+- You may change the Bert size by passing a different [checkpoint](https://huggingface.co/transformers/pretrained_models.html) to the `--model_name_or_path` argument.
+
+- Passing your user name to the `--wandb_user_name` argument will trigger weights and biases logging.
+
+- You can choose a dataset with `--dataset_name` and `--dataset_config`. Our [viewer](https://huggingface.co/datasets/viewer/) will help you find what you need.
\ No newline at end of file

examples/research_projects/performer/full_script.sh

+TOKENIZERS_PARALLELISM=true python run_mlm_performer.py  --output_dir experiments --dataset_name wikipedia --dataset_config_name 20200501.en --model_name_or_path bert-large-cased --tokenizer_name bert-large-cased --do_train --overwrite_output_dir --per_device_train_batch_size 4 --learning_rate 5e-4 --warmup_steps 100 --num_train_epochs 3 --performer
\ No newline at end of file

examples/research_projects/performer/sanity_script.sh

+TOKENIZERS_PARALLELISM=true python run_mlm_performer.py  --output_dir experiments --dataset_name wikipedia --dataset_config_name 20200501.simple --model_name_or_path bert-base-cased --tokenizer_name bert-base-cased --do_train --overwrite_output_dir --per_device_train_batch_size 4 --learning_rate 5e-4 --warmup_steps 100 --num_train_epochs 3 --performer
\ No newline at end of file