Reducing training and validation steps (huggingface#8)

Author: Steven I Reeves

scripts/gpt2-tf2/gpt2_train_distributed.py

@@ -50,7 +50,7 @@ def get_dataset(fil):
 
 def tokenize(data, tokenizer, truncate=False):
     if truncate:
-        data = tokenizer(data[:1000], return_tensors='tf', padding=True, truncation=True)
+        data = tokenizer(data[:100], return_tensors='tf', padding=True, truncation=True)
     else:
         data = tokenizer(data, return_tensors='tf', padding=True, truncation=True)
     return tf.data.Dataset.from_tensor_slices((dict(data), data['input_ids']))