Create modelcard if doesn't exist on update_metadata (#1061)

* Create modelcard if not exist on update_metadata

* fix test

* fix unicode characters in modelcards

* unicode by default in yaml

* Add test for update metadata on missing repocard

* docstring

* remove useless comment

Author: Wauplin

src/huggingface_hub/keras_mixin.py

@@ -8,14 +8,14 @@
 from typing import Any, Dict, List, Optional, Union
 from urllib.parse import quote
 
-import yaml
 from huggingface_hub import CommitOperationDelete, ModelHubMixin, snapshot_download
 from huggingface_hub.file_download import (
     get_tf_version,
     is_graphviz_available,
     is_pydot_available,
     is_tf_available,
 )
+from huggingface_hub.utils import yaml_dump
 
 from .constants import CONFIG_NAME, DEFAULT_REVISION
 from .hf_api import HfApi, _parse_revision_from_pr_url, _prepare_upload_folder_commit
@@ -105,7 +105,7 @@ def _create_model_card(
     readme_path = f"{repo_dir}/README.md"
     metadata["library_name"] = "keras"
     model_card = "---\n"
-    model_card += yaml.dump(metadata, default_flow_style=False)
+    model_card += yaml_dump(metadata, default_flow_style=False)
     model_card += "---\n"
     model_card += "\n## Model description\n\nMore information needed\n"
     model_card += "\n## Intended uses & limitations\n\nMore information needed\n"

src/huggingface_hub/repocard.py

@@ -3,7 +3,7 @@
 import sys
 import tempfile
 from pathlib import Path
-from typing import Any, Dict, Optional, Union
+from typing import Any, Dict, Optional, Type, Union
 
 
 if sys.version_info >= (3, 8):
@@ -23,9 +23,10 @@
     eval_results_to_model_index,
     model_index_to_eval_results,
 )
+from huggingface_hub.utils import yaml_dump
 
 from .constants import REPOCARD_NAME
-from .utils import validate_hf_hub_args
+from .utils import EntryNotFoundError, validate_hf_hub_args
 from .utils.logging import get_logger
 
 
@@ -512,7 +513,7 @@ def metadata_save(local_path: Union[str, Path], data: Dict) -> None:
 
     # creates a new file if it not
     with open(local_path, "w", newline="") as readme:
-        data_yaml = yaml.dump(data, sort_keys=False, line_break=line_break)
+        data_yaml = yaml_dump(data, sort_keys=False, line_break=line_break)
         # sort_keys: keep dict order
         match = REGEX_YAML_BLOCK.search(content)
         if match:
@@ -669,6 +670,9 @@ def metadata_update(
 ) -> str:
     """
     Updates the metadata in the README.md of a repository on the Hugging Face Hub.
+    If the README.md file doesn't exist yet, a new one is created with metadata and an
+    the default ModelCard or DatasetCard template. For `space` repo, an error is thrown
+    as a Space cannot exist without a `README.md` file.
 
     Args:
         repo_id (`str`):
@@ -685,7 +689,7 @@ def metadata_update(
             The Hugging Face authentication token.
         commit_message (`str`, *optional*):
             The summary / title / first line of the generated commit. Defaults to
-            `f"Update metdata with huggingface_hub"`
+            `f"Update metadata with huggingface_hub"`
         commit_description (`str` *optional*)
             The description of the generated commit
         revision (`str`, *optional*):
@@ -724,7 +728,30 @@ def metadata_update(
         else "Update metadata with huggingface_hub"
     )
 
-    card = ModelCard.load(repo_id, token=token)
+    # Card class given repo_type
+    card_class: Type[RepoCard]
+    if repo_type is None or repo_type == "model":
+        card_class = ModelCard
+    elif repo_type == "dataset":
+        card_class = DatasetCard
+    elif repo_type == "space":
+        card_class = RepoCard
+    else:
+        raise ValueError(f"Unknown repo_type: {repo_type}")
+
+    # Either load repo_card from the Hub or create an empty one.
+    # NOTE: Will not create the repo if it doesn't exist.
+    try:
+        card = card_class.load(repo_id, token=token, repo_type=repo_type)
+    except EntryNotFoundError:
+        if repo_type == "space":
+            raise ValueError(
+                "Cannot update metadata on a Space that doesn't contain a `README.md`"
+                " file."
+            )
+
+        # Initialize a ModelCard or DatasetCard from default template and no data.
+        card = card_class.from_template(CardData())
 
     for key, value in metadata.items():
         if key == "model-index":

src/huggingface_hub/repocard_data.py

@@ -3,7 +3,7 @@
 from dataclasses import dataclass
 from typing import Any, Dict, List, Optional, Tuple, Union
 
-import yaml
+from huggingface_hub.utils import yaml_dump
 
 from .utils.logging import get_logger
 
@@ -152,7 +152,7 @@ def to_yaml(self, line_break=None) -> str:
         Returns:
             `str`: CardData represented as a YAML block.
         """
-        return yaml.dump(self.to_dict(), sort_keys=False, line_break=line_break).strip()
+        return yaml_dump(self.to_dict(), sort_keys=False, line_break=line_break).strip()
 
     def __repr__(self):
         return self.to_yaml()

src/huggingface_hub/utils/__init__.py

@@ -35,6 +35,7 @@
     RevisionNotFoundError,
     hf_raise_for_status,
 )
+from ._fixes import yaml_dump
 from ._headers import build_hf_headers
 from ._hf_folder import HfFolder
 from ._http import http_backoff

src/huggingface_hub/utils/_fixes.py

@@ -8,3 +8,20 @@
         from simplejson import JSONDecodeError  # noqa
     except ImportError:
         from json import JSONDecodeError  # noqa
+
+from functools import partial
+
+import yaml
+
+
+# Wrap `yaml.dump` to set `allow_unicode=True` by default.
+#
+# Example:
+# ```py
+# >>> yaml.dump({"emoji": "👀", "some unicode": "日本か"})
+# 'emoji: "\\U0001F440"\nsome unicode: "\\u65E5\\u672C\\u304B"\n'
+#
+# >>> yaml_dump({"emoji": "👀", "some unicode": "日本か"})
+# 'emoji: "👀"\nsome unicode: "日本か"\n'
+# ```
+yaml_dump = partial(yaml.dump, allow_unicode=True)

tests/test_repocard.py

@@ -418,6 +418,61 @@ def test_update_metadata_on_empty_text_content(self) -> None:
         self.assertDictEqual(updated_metadata, expected_metadata)
 
 
+class TestMetadataUpdateOnMissingCard(unittest.TestCase):
+    def setUp(self) -> None:
+        """
+        Share this valid token in all tests below.
+        """
+        self._token = TOKEN
+        self._api = HfApi(endpoint=ENDPOINT_STAGING)
+        self._repo_id = f"{USER}/{repo_name()}"
+
+    def test_metadata_update_missing_readme_on_model(self) -> None:
+        self._api.create_repo(self._repo_id, token=self._token)
+        metadata_update(self._repo_id, {"tag": "this_is_a_test"}, token=self._token)
+        model_card = ModelCard.load(self._repo_id, token=self._token)
+
+        # Created a card with default template + metadata
+        self.assertIn("# Model Card for Model ID", str(model_card))
+        self.assertEqual(model_card.data.to_dict(), {"tag": "this_is_a_test"})
+
+        self._api.delete_repo(self._repo_id, token=self._token)
+
+    def test_metadata_update_missing_readme_on_dataset(self) -> None:
+        self._api.create_repo(self._repo_id, repo_type="dataset", token=self._token)
+        metadata_update(
+            self._repo_id,
+            {"tag": "this is a dataset test"},
+            token=self._token,
+            repo_type="dataset",
+        )
+        dataset_card = DatasetCard.load(self._repo_id, token=self._token)
+
+        # Created a card with default template + metadata
+        self.assertIn("# Dataset Card for Dataset Name", str(dataset_card))
+        self.assertEqual(dataset_card.data.to_dict(), {"tag": "this is a dataset test"})
+
+        self._api.delete_repo(self._repo_id, repo_type="dataset", token=self._token)
+
+    def test_metadata_update_missing_readme_on_space(self) -> None:
+        self._api.create_repo(
+            self._repo_id, repo_type="space", token=self._token, space_sdk="static"
+        )
+        self._api.delete_file(
+            "README.md", self._repo_id, repo_type="space", token=self._token
+        )
+        with self.assertRaises(ValueError):
+            # Cannot create a default readme on a space repo (should be automatically
+            # created on the Hub).
+            metadata_update(
+                self._repo_id,
+                {"tag": "this is a space test"},
+                token=self._token,
+                repo_type="space",
+            )
+        self._api.delete_repo(self._repo_id, repo_type="space", token=self._token)
+
+
 class TestCaseWithCapLog(unittest.TestCase):
     _api = HfApi(endpoint=ENDPOINT_STAGING)
 

tests/test_utils_fixes.py

@@ -0,0 +1,16 @@
+import unittest
+
+from huggingface_hub.utils import yaml_dump
+
+
+class TestYamlDump(unittest.TestCase):
+    def test_yaml_dump_emoji(self) -> None:
+        self.assertEqual(yaml_dump({"emoji": "👀"}), "emoji: 👀\n")
+
+    def test_yaml_dump_japanese_characters(self) -> None:
+        self.assertEqual(yaml_dump({"some unicode": "日本か"}), "some unicode: 日本か\n")
+
+    def test_yaml_dump_explicit_no_unicode(self) -> None:
+        self.assertEqual(
+            yaml_dump({"emoji": "👀"}, allow_unicode=False), 'emoji: "\\U0001F440"\n'
+        )