tests: switch/fix XSum test dataset (#310)

* switch test dataset
* fix CNN
* workers
* chlog

CHANGELOG.md

@@ -5,6 +5,13 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 
+## [0.2.5] - 2022-11-DD
+
+### Fixed
+
+- Fixed passing config name to `CNNDailyMailSummarizationDataModule` ([#310](https://github.com/Lightning-AI/lightning-transformers/pull/310))
+
+
 ## [0.2.4] - 2022-11-03
 
 ### Changed

lightning_transformers/task/nlp/summarization/datasets/cnn_dailymail.py

@@ -17,8 +17,8 @@
 
 
 class CNNDailyMailSummarizationDataModule(SummarizationDataModule):
-    def __init__(self, *args, dataset_name: str = "cnn_dailymail", **kwargs):
-        super().__init__(*args, dataset_name=dataset_name, **kwargs)
+    def __init__(self, *args, dataset_name: str = "cnn_dailymail", config_name: str = "3.0.0", **kwargs):
+        super().__init__(*args, dataset_name=dataset_name, dataset_config_name=config_name, **kwargs)
 
     @property
     def source_target_column_names(self) -> Tuple[str, str]:

tests/task/nlp/test_summarization.py

@@ -7,27 +7,30 @@
 from transformers import AutoTokenizer
 
 from lightning_transformers.task.nlp.summarization import (
+    CNNDailyMailSummarizationDataModule,
     SummarizationDataModule,
     SummarizationTransformer,
-    XsumSummarizationDataModule,
 )
 
+_MODEL_TINY = "patrickvonplaten/t5-tiny-random"
+
 
 @pytest.mark.skipif(sys.platform == "win32", reason="Currently Windows is not supported")
 def test_smoke_train(hf_cache_path):
-    tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path="patrickvonplaten/t5-tiny-random")
+    tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path=_MODEL_TINY)
     model = SummarizationTransformer(
-        pretrained_model_name_or_path="patrickvonplaten/t5-tiny-random",
+        pretrained_model_name_or_path=_MODEL_TINY,
         use_stemmer=True,
         val_target_max_length=142,
         num_beams=None,
         compute_generate_metrics=True,
     )
-    dm = XsumSummarizationDataModule(
+    dm = CNNDailyMailSummarizationDataModule(
         limit_train_samples=64,
         limit_val_samples=64,
         limit_test_samples=64,
-        batch_size=1,
+        num_workers=2,
+        preprocessing_num_workers=2,
         max_source_length=128,
         max_target_length=128,
         cache_dir=hf_cache_path,
@@ -41,8 +44,8 @@ def test_smoke_train(hf_cache_path):
 @pytest.mark.skipif(sys.platform == "win32", reason="Currently Windows is not supported")
 def test_smoke_predict():
     model = SummarizationTransformer(
-        pretrained_model_name_or_path="patrickvonplaten/t5-tiny-random",
-        tokenizer=AutoTokenizer.from_pretrained(pretrained_model_name_or_path="patrickvonplaten/t5-tiny-random"),
+        pretrained_model_name_or_path=_MODEL_TINY,
+        tokenizer=AutoTokenizer.from_pretrained(pretrained_model_name_or_path=_MODEL_TINY),
     )
 
     y = model.hf_predict(