[docs] Add integration test example to copy pasta template (#5961)

Co-authored-by: Julien Chaumond <chaumond@gmail.com>

templates/adding_a_new_example_script/README.md

@@ -1,5 +1,15 @@
 # How to add a new example script in 🤗Transformers
 
 This folder provide a template for adding a new example script implementing a training or inference task with the models in the  🤗Transformers library.
+Add tests!
 
-Currently only examples for PyTorch are provided which are adaptations of the library's SQuAD examples which implement single-GPU and distributed training with gradient accumulation and mixed-precision (using NVIDIA's apex library) to cover a reasonable range of use cases.
+
+These folder can be put in a subdirectory under your example's name, like `examples/deebert`.
+
+
+Best Practices: 
+- use `Trainer`/`TFTrainer`
+- write an @slow test that checks that your model can train on one batch and get a low loss.
+    - this test should use cuda if it's available. (e.g. by checking `transformers.torch_device`)
+- adding an `eval_xxx.py` script that can evaluate a pretrained checkpoint.  
+- tweet about your new example with a carbon screenshot of how to run it and tag @huggingface