Add cache_dir to save features in GLUE + Differentiate match/mismatch for MNLI metrics (#4621)

* Glue task cleaup

* Enable writing cache to cache_dir in case dataset lives in readOnly
filesystem.
* Differentiate match vs mismatch for MNLI metrics.

* Style

* Fix pytype

* Fix type

* Use cache_dir in mnli mismatch eval dataset

* Small Tweaks

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

src/transformers/data/datasets/glue.py

@@ -70,6 +70,7 @@ class GlueDataset(Dataset):
         tokenizer: PreTrainedTokenizer,
         limit_length: Optional[int] = None,
         mode: Union[str, Split] = Split.train,
+        cache_dir: Optional[str] = None,
     ):
         self.args = args
         self.processor = glue_processors[args.task_name]()
@@ -81,7 +82,7 @@ class GlueDataset(Dataset):
                 raise KeyError("mode is not a valid split name")
         # Load data features from cache or dataset file
         cached_features_file = os.path.join(
-            args.data_dir,
+            cache_dir if cache_dir is not None else args.data_dir,
             "cached_{}_{}_{}_{}".format(
                 mode.value, tokenizer.__class__.__name__, str(args.max_seq_length), args.task_name,
             ),

src/transformers/data/metrics/__init__.py

@@ -63,9 +63,9 @@ if _has_sklearn:
         elif task_name == "qqp":
             return acc_and_f1(preds, labels)
         elif task_name == "mnli":
-            return {"acc": simple_accuracy(preds, labels)}
+            return {"mnli/acc": simple_accuracy(preds, labels)}
         elif task_name == "mnli-mm":
-            return {"acc": simple_accuracy(preds, labels)}
+            return {"mnli-mm/acc": simple_accuracy(preds, labels)}
         elif task_name == "qnli":
             return {"acc": simple_accuracy(preds, labels)}
         elif task_name == "rte":