Fixing issue where generic model types wouldn't load properly with the pipeline

src/transformers/pipelines/__init__.py

@@ -667,6 +667,15 @@ def pipeline(
         # the files could be missing from the hub, instead of failing
         # on such repos, we just force to not load it.
         load_tokenizer = False
+
+    if not load_feature_extractor and task not in NO_FEATURE_EXTRACTOR_TASKS:
+        raise EnvironmentError(
+            f"There is a problem in `transformers`. The task {task} requires a feature extractor, however the model {type(model_config)} seems to not support feature-extractors. Please report this issue. This is likely a misconfiguration in the library, please report this issue."
+        )
+    if not load_tokenizer and task not in NO_TOKENIZER_TASKS:
+        raise EnvironmentError(
+            f"There is a problem in `transformers`. The task {task} requires a tokenizer, however the model {type(model_config)} seems to not support tokenizer. This is likely a misconfiguration in the library, please report this issue."
+        )
     if task in NO_FEATURE_EXTRACTOR_TASKS:
         load_feature_extractor = False
 

tests/pipelines/test_pipelines_zero_shot_image_classification.py

@@ -172,6 +172,22 @@ def test_small_model_tf(self):
             ],
         )
 
+    @slow
+    @require_torch
+    def test_large_model_misconfigured(self):
+        # XXX this should be a fast test, but the triggering arch
+        # VisionTextDualEncoderModel is missing for small tests
+        # https://huggingface.co/hf-internal-testing
+        # This test will also start to fail, once this architecture
+        # correctly defines AutoFeatureExtractor. At this point
+        # we can safely remove this test as we don't really want
+        # to keep around an invalid model around just for this.
+        with self.assertRaises(EnvironmentError):
+            pipeline(
+                task="zero-shot-image-classification",
+                model="Bingsu/vitB32_bert_ko_small_clip",
+            )
+
     @slow
     @require_torch
     def test_large_model_pt(self):