make tests order independent

.github/trigger_files/beam_PostCommit_Python_Dependency.json

@@ -1,4 +1,4 @@
 {
     "comment": "Modify this file in a trivial way to cause this test suite to run",
-    "modification": 9
+    "modification": 13
   }

sdks/python/apache_beam/examples/snippets/transforms/elementwise/mltransform_test.py

@@ -72,7 +72,8 @@ def test_mltransform_compute_and_apply_vocab(self, mock_stdout):
     mltransform_compute_and_apply_vocabulary()
     predicted = mock_stdout.getvalue().splitlines()
     expected = check_mltransform_compute_and_apply_vocab()
-    self.assertEqual(predicted, expected)
+    # Print order from the pipeline is not guaranteed.
+    self.assertCountEqual(predicted, expected)
 
   def test_mltransform_scale_to_0_1(self, mock_stdout):
     mltransform_scale_to_0_1()
@@ -84,7 +85,8 @@ def test_mltransform_compute_and_apply_vocab_scalar(self, mock_stdout):
     mltransform_compute_and_apply_vocabulary_with_scalar()
     predicted = mock_stdout.getvalue().splitlines()
     expected = check_mltransform_compute_and_apply_vocabulary_with_scalar()
-    self.assertEqual(predicted, expected)
+    # Print order from the pipeline is not guaranteed.
+    self.assertCountEqual(predicted, expected)
 
 
 if __name__ == '__main__':

sdks/python/apache_beam/ml/transforms/embeddings/huggingface.py

@@ -215,7 +215,14 @@ def run_inference(
         self._config.api_url,
         headers=self._config.authorization_token,
         json={"inputs": batch})
-    return response.json()
+    response.raise_for_status()
+    result = response.json()
+    # Reject {"error": ...} so it is not treated as embedding values.
+    if isinstance(result, dict) and 'error' in result:
+      raise RuntimeError(
+          'Hugging Face Inference API error for %s: %s' %
+          (self._config.api_url, result['error']))
+    return result
 
 
 class InferenceAPIEmbeddings(EmbeddingsManager):

sdks/python/apache_beam/ml/transforms/embeddings/huggingface_test.py

@@ -27,6 +27,7 @@
 import jsonpickle
 import numpy as np
 import pytest
+import requests
 from parameterized import parameterized
 
 import apache_beam as beam
@@ -70,7 +71,33 @@
 # InferenceAPIEmbeddings calls /pipeline/feature-extraction.
 INFERENCE_API_MODEL_NAME = "BAAI/bge-small-en-v1.5"
 INFERENCE_API_EMBEDDING_DIM = 384
+INFERENCE_API_URL = (
+    'https://router.huggingface.co/hf-inference/models/'
+    f'{INFERENCE_API_MODEL_NAME}/pipeline/feature-extraction')
 IMAGE_MODEL_NAME = "clip-ViT-B-32"
+
+
+def _hf_inference_api_available():
+  """Returns (ready, reason) for the optional live Inference API dependency."""
+  if not _HF_TOKEN:
+    return False, 'HF_INFERENCE_TOKEN environment variable not set.'
+  try:
+    response = requests.post(
+        INFERENCE_API_URL,
+        headers={'Authorization': f'Bearer {_HF_TOKEN}'},
+        json={'inputs': test_query},
+        timeout=60)
+    payload = response.json()
+  except Exception as exc:  # pylint: disable=broad-except
+    return False, 'HF Inference API unavailable: %s' % exc
+  if response.status_code >= 400:
+    return False, 'HF Inference API HTTP %s: %s' % (
+        response.status_code, payload)
+  if isinstance(payload, dict) and 'error' in payload:
+    return False, 'HF Inference API error: %s' % payload['error']
+  return True, ''
+
+
 _parameterized_inputs = [
     ([{
         test_query_column: 'That is a happy person'
@@ -476,7 +503,8 @@ def test_loading_model_from_artifacts_written_by_older_versions(self):
     self.assertEqual(model.model_name, DEFAULT_MODEL_NAME)
 
 
-@unittest.skipIf(not _HF_TOKEN, 'HF_TOKEN environment variable not set.')
+@unittest.skipIf(
+    not _HF_TOKEN, 'HF_INFERENCE_TOKEN environment variable not set.')
 class HuggingfaceInferenceAPITest(unittest.TestCase):
   def setUp(self):
     self.artifact_location = tempfile.mkdtemp()
@@ -494,6 +522,9 @@ def test_get_api_url_and_when_model_name_not_provided(self):
       _ = inference_embeddings.api_url
 
   def test_embeddings_with_inference_api(self):
+    ready, reason = _hf_inference_api_available()
+    if not ready:
+      self.skipTest(reason)
     embedding_config = InferenceAPIEmbeddings(
         hf_token=_HF_TOKEN,
         model_name=INFERENCE_API_MODEL_NAME,
@@ -516,7 +547,8 @@ def _embedding_shape(element):
           equal_to([(INFERENCE_API_EMBEDDING_DIM, True)]))
 
 
-@unittest.skipIf(not _HF_TOKEN, 'HF_TOKEN environment variable not set.')
+@unittest.skipIf(
+    not _HF_TOKEN, 'HF_INFERENCE_TOKEN environment variable not set.')
 class HuggingfaceInferenceAPIGCSLocationTest(HuggingfaceInferenceAPITest):
   def setUp(self):
     self.artifact_location = self.gcs_artifact_location = os.path.join(