From 6f8f2f6a77457f993a582bde2bff92af863a2d06 Mon Sep 17 00:00:00 2001
From: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Date: Wed, 14 Sep 2022 07:36:12 -0400
Subject: [PATCH] Make AutoProcessor a magic loading class for all modalities
 (#18963)

* Make AutoProcessor a magic loading class for all modalities

* Quality
---
 .../models/auto/processing_auto.py            | 24 +++++++++++++++----
 tests/models/auto/test_processor_auto.py      |  8 +++++++
 2 files changed, 28 insertions(+), 4 deletions(-)

diff --git a/src/transformers/models/auto/processing_auto.py b/src/transformers/models/auto/processing_auto.py
index 7eff84c5d5..07b2811a16 100644
--- a/src/transformers/models/auto/processing_auto.py
+++ b/src/transformers/models/auto/processing_auto.py
@@ -23,7 +23,7 @@ from ...configuration_utils import PretrainedConfig
 from ...dynamic_module_utils import get_class_from_dynamic_module
 from ...feature_extraction_utils import FeatureExtractionMixin
 from ...tokenization_utils import TOKENIZER_CONFIG_FILE
-from ...utils import CONFIG_NAME, FEATURE_EXTRACTOR_NAME, get_file_from_repo, logging
+from ...utils import FEATURE_EXTRACTOR_NAME, get_file_from_repo, logging
 from .auto_factory import _LazyAutoMapping
 from .configuration_auto import (
     CONFIG_MAPPING_NAMES,
@@ -31,6 +31,8 @@ from .configuration_auto import (
     model_type_to_module_name,
     replace_list_option_in_docstrings,
 )
+from .feature_extraction_auto import AutoFeatureExtractor
+from .tokenization_auto import AutoTokenizer
 
 
 logger = logging.get_logger(__name__)
@@ -250,10 +252,24 @@ class AutoProcessor:
         if type(config) in PROCESSOR_MAPPING:
             return PROCESSOR_MAPPING[type(config)].from_pretrained(pretrained_model_name_or_path, **kwargs)
 
+        # At this stage, there doesn't seem to be a `Processor` class available for this model, so let's try a
+        # tokenizer.
+        try:
+            return AutoTokenizer.from_pretrained(
+                pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
+            )
+        except Exception:
+            try:
+                return AutoFeatureExtractor.from_pretrained(
+                    pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
+                )
+            except Exception:
+                pass
+
         raise ValueError(
-            f"Unrecognized processor in {pretrained_model_name_or_path}. Should have a `processor_type` key in "
-            f"its {FEATURE_EXTRACTOR_NAME}, or one of the following `model_type` keys in its {CONFIG_NAME}: "
-            f"{', '.join(c for c in PROCESSOR_MAPPING_NAMES.keys())}"
+            f"Unrecognized processing class in {pretrained_model_name_or_path}. Can't instantiate a processor, a "
+            "tokenizer or a feature extractor for this model. Make sure the repository contains the files of at least "
+            "one of those processing classes."
         )
 
     @staticmethod
diff --git a/tests/models/auto/test_processor_auto.py b/tests/models/auto/test_processor_auto.py
index 2f99d5c379..fe57078a61 100644
--- a/tests/models/auto/test_processor_auto.py
+++ b/tests/models/auto/test_processor_auto.py
@@ -202,6 +202,14 @@ class AutoFeatureExtractorTest(unittest.TestCase):
             if CustomConfig in PROCESSOR_MAPPING._extra_content:
                 del PROCESSOR_MAPPING._extra_content[CustomConfig]
 
+    def test_auto_processor_creates_tokenizer(self):
+        processor = AutoProcessor.from_pretrained("hf-internal-testing/tiny-random-bert")
+        self.assertEqual(processor.__class__.__name__, "BertTokenizerFast")
+
+    def test_auto_processor_creates_feature_extractor(self):
+        processor = AutoProcessor.from_pretrained("hf-internal-testing/tiny-random-convnext")
+        self.assertEqual(processor.__class__.__name__, "ConvNextFeatureExtractor")
+
 
 @is_staging_test
 class ProcessorPushToHubTester(unittest.TestCase):