BIG Reorganize examples (#4213)

* Created using Colaboratory

* [examples] reorganize files

* remove run_tpu_glue.py as superseded by TPU support in Trainer

* Bugfix: int, not tuple

* move files around

.gitignore

@@ -132,7 +132,8 @@ proc_data
 runs
 /runs_old
 /wandb
-examples/runs
+/examples/runs
+/examples/**/*.args
 
 # data
 /data

README.md

@@ -331,7 +331,7 @@ pip install -r ./examples/requirements.txt
 export GLUE_DIR=/path/to/glue
 export TASK_NAME=MRPC
 
-python ./examples/run_glue.py \
+python ./examples/text-classification/run_glue.py \
     --model_name_or_path bert-base-uncased \
     --task_name $TASK_NAME \
     --do_train \
@@ -357,7 +357,7 @@ Parallel training is a simple way to use several GPUs (but is slower and less fl
 ```shell
 export GLUE_DIR=/path/to/glue
 
-python ./examples/run_glue.py \
+python ./examples/text-classification/run_glue.py \
     --model_name_or_path xlnet-large-cased \
     --do_train  \
     --do_eval   \
@@ -382,7 +382,7 @@ On this machine we thus have a batch size of 32, please increase `gradient_accum
 This example code fine-tunes the Bert Whole Word Masking model on the Microsoft Research Paraphrase Corpus (MRPC) corpus using distributed training on 8 V100 GPUs to reach a F1 > 92.
 
 ```bash
-python -m torch.distributed.launch --nproc_per_node 8 ./examples/run_glue.py   \
+python -m torch.distributed.launch --nproc_per_node 8 ./examples/text-classification/run_glue.py   \
     --model_name_or_path bert-large-uncased-whole-word-masking \
     --task_name MRPC \
     --do_train   \