[trainer] parametrize default output_dir

This PR:

* fixes trainer to have the logger agree with the actual default `output_dir`, but setting it one place and passing it as an argument to both places

@sgugger

examples/language-modeling/README.md

@@ -30,15 +30,6 @@ need extra processing on your datasets.
 The following examples, will run on a datasets hosted on our [hub](https://huggingface.co/datasets) or with your own
 text files for training and validation. We give examples of both below.
 
-### Datasets
-
-The main dataset used in the examples below is:
-
-```bash
-wget https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-raw-v1.zip
-unzip wikitext-2-raw-v1.zip
-```
-
 ### GPT-2/GPT and causal language modeling
 
 The following example fine-tunes GPT-2 on WikiText-2. We're using the raw WikiText-2 (no tokens were replaced before

src/transformers/trainer.py

@@ -228,8 +228,9 @@ class Trainer:
         optimizers: Tuple[torch.optim.Optimizer, torch.optim.lr_scheduler.LambdaLR] = (None, None),
     ):
         if args is None:
-            logger.info("No `TrainingArguments` passed, using the current path as `output_dir`.")
-            args = TrainingArguments("tmp_trainer")
+            output_dir = "tmp_trainer"
+            logger.info(f"No `TrainingArguments` passed, using `output_dir={output_dir}`.")
+            args = TrainingArguments(output_dir=output_dir)
         self.args = args
         # Seed must be set before instantiating the model when using model
         set_seed(self.args.seed)