Allow '#' in the task name. This will be used as a separator for deno…

…ting different versions of the same task. PiperOrigin-RevId: 612779694
google · Mar 5, 2024 · 11706e4 · 11706e4
1 parent 16380a6
commit 11706e4
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 1 deletion.
diff --git a/seqio/dataset_providers.py b/seqio/dataset_providers.py
@@ -52,7 +52,7 @@
 
 _DEFAULT_FEATURE_KEYS = ["inputs", "targets"]
 
-_VALID_TASK_NAME_REGEX = re.compile(r"^[\w\d\.\:_]+$")
+_VALID_TASK_NAME_REGEX = re.compile(r"^[\w\d\.\:_#]+$")
 _MAX_EXAMPLES_TO_MEM_CACHE = 10000
 SHUFFLE_BUFFER_SIZE = 1000
 

diff --git a/seqio/dataset_providers_test.py b/seqio/dataset_providers_test.py
@@ -59,6 +59,9 @@ def test_invalid_name(self):
     ):
       self.add_task("invalid/name", self.function_source)
 
+  def test_hash_in_name(self):
+    self.add_task("valid#name", self.function_source)
+
   def test_repeat_name(self):
     with self.assertRaisesWithLiteralMatch(
         ValueError, "Attempting to register duplicate provider: text_line_task"