[s2s] distill t5-large -> t5-small (#8376)

Co-authored-by: Sam Shleifer <sshleifer@gmail.com>

examples/seq2seq/test_seq2seq_examples.py

@@ -96,6 +96,7 @@ CHEAP_ARGS = {
     "freeze_encoder": False,
     "auto_scale_batch_size": False,
     "overwrite_output_dir": False,
+    "student": None,
 }
 
 
@@ -107,6 +108,7 @@ def _dump_articles(path: Path, articles: list):
 ARTICLES = [" Sam ate lunch today.", "Sams lunch ingredients."]
 SUMMARIES = ["A very interesting story about what I ate for lunch.", "Avocado, celery, turkey, coffee"]
 T5_TINY = "patrickvonplaten/t5-tiny-random"
+T5_TINIER = "sshleifer/t5-tinier-random"
 BART_TINY = "sshleifer/bart-tiny-random"
 MBART_TINY = "sshleifer/tiny-mbart"
 MARIAN_TINY = "sshleifer/tiny-marian-en-de"
@@ -239,6 +241,16 @@ class TestSummarizationDistiller(TestCasePlus):
         )
         self._test_distiller_cli(updates)
 
+    @require_torch_non_multi_gpu_but_fix_me
+    def test_distill_different_base_models(self):
+        updates = dict(
+            teacher=T5_TINY,
+            student=T5_TINIER,
+            model_name_or_path=T5_TINIER,
+            tokenizer_name=T5_TINIER,
+        )
+        self._test_distiller_cli(updates)
+
     def _test_distiller_cli(self, updates, check_contents=True):
         default_updates = dict(
             label_smoothing=0.0,