fix: stop grouping HF/ModelScope models by repository

A repository is not a model identity: collection repos on Hugging Face
and ModelScope host many unrelated models, which were wrongly shown as
versions of each other.

- Hugging Face models no longer auto-group (the Hub exposes no
  site-native model id)
- ModelScope models group by the site's native published-model id
  (MuseInfo modelVersion.modelId), extracted during enrichment and
  persisted on the sidecar as source_model_id/source_version_id;
  unenriched models stay standalone instead of collapsing a whole repo
  into one group
- TensorArt grouping unchanged (its URL id is already model-level)
- Frontend group-key derivation mirrors the new backend semantics

tests/frontend/utils/modelSourceHelpers.test.js

@@ -105,13 +105,39 @@ describe('modelSourceHelpers', () => {
 
   describe('getModelSourceGroupKey', () => {
     it('matches the backend group-key shapes', () => {
-      expect(getModelSourceGroupKey({ hf_url: 'https://huggingface.co/u/r' })).toBe('hf:u/r');
-      expect(
-        getModelSourceGroupKey({ source_url: 'https://modelscope.cn/models/u/r' })
-      ).toBe('ms:u/r');
+      // TensorArt's numeric id already identifies a single model.
       expect(getModelSourceGroupKey({ source_url: 'https://tensor.art/models/123' })).toBe(
         'ta:123'
       );
+      // ModelScope groups by the site-native published-model id.
+      expect(
+        getModelSourceGroupKey({
+          source_url: 'https://modelscope.cn/models/u/r',
+          source_model_id: '555',
+        })
+      ).toBe('ms:555');
+      expect(
+        getModelSourceGroupKey({
+          source_url: 'https://www.modelscope.ai/models/u/r',
+          source_model_id: '678',
+        })
+      ).toBe('msai:678');
+    });
+
+    it('returns an empty string for sources without a model identity', () => {
+      // Hugging Face repos are not a model identity: never grouped.
+      expect(getModelSourceGroupKey({ hf_url: 'https://huggingface.co/u/r' })).toBe('');
+      // Unenriched ModelScope models stay standalone rather than collapsing
+      // a whole collection repo into one group.
+      expect(
+        getModelSourceGroupKey({ source_url: 'https://modelscope.cn/models/u/r' })
+      ).toBe('');
+      expect(
+        getModelSourceGroupKey({
+          source_url: 'https://modelscope.cn/models/u/r',
+          source_model_id: '   ',
+        })
+      ).toBe('');
     });
 
     it('returns an empty string without a source', () => {

tests/routes/test_model_source_handlers.py

@@ -1027,6 +1027,8 @@ def _modelscope_card_payload() -> dict:
                         "modelVersion": {
                             "showName": "c1-st1000",
                             "triggerWords": '["kreaface","kreamodel"]',
+                            "id": 1002,
+                            "modelId": 555,
                         },
                         "coverImages": [
                             {"url": "https://resources.modelscope.cn/cover-images/b.png"},
@@ -1141,6 +1143,9 @@ async def test_download_hydrates_the_card_from_the_site(tmp_path, monkeypatch):
         '{"strength_min": 0.5, "strength_max": 1.2, "strength_range": "0.5-1.2"}'
     )
     assert saved["metadata_source"] == "source:modelscope"
+    # The site-native identity ids are persisted for version grouping.
+    assert saved["source_model_id"] == "555"
+    assert saved["source_version_id"] == "1002"
     # No provider answered, so claiming an AI enrichment would be a lie.
     assert "llm_enriched_at" not in saved
 
@@ -1148,6 +1153,7 @@ async def test_download_hydrates_the_card_from_the_site(tmp_path, monkeypatch):
     assert scanner.update_single_model_cache.await_count == 1
     cached = scanner.update_single_model_cache.await_args.args[2]
     assert cached["model_name"] == "Krea-2-LORA"
+    assert cached["source_model_id"] == "555"
 
 
 @pytest.mark.asyncio

tests/services/test_base_model_service.py

@@ -1263,39 +1263,8 @@ async def test_get_model_civitai_url_falls_back_when_host_setting_is_not_a_strin
     }
 
 
-class TestHfGroupKey:
-    """Tests for _extract_hf_group_key and _extract_group_key."""
-
-    # --- _extract_hf_group_key ---
-
-    def test_hf_group_key_valid_url(self):
-        """Standard HF URL returns hf:user/repo."""
-        item = {"hf_url": "https://huggingface.co/unsloth/qwen-edit"}
-        assert BaseModelService._extract_hf_group_key(item) == "hf:unsloth/qwen-edit"
-
-    def test_hf_group_key_url_with_subpath(self):
-        """URL with subpath still extracts just owner/repo."""
-        item = {"hf_url": "https://huggingface.co/user/repo/resolve/main/file.safetensors"}
-        assert BaseModelService._extract_hf_group_key(item) == "hf:user/repo"
-
-    def test_hf_group_key_empty_url(self):
-        """Empty hf_url returns None."""
-        assert BaseModelService._extract_hf_group_key({"hf_url": ""}) is None
-
-    def test_hf_group_key_no_url(self):
-        """Missing hf_url key returns None."""
-        assert BaseModelService._extract_hf_group_key({}) is None
-
-    def test_hf_group_key_none_url(self):
-        """None hf_url returns None."""
-        assert BaseModelService._extract_hf_group_key({"hf_url": None}) is None
-
-    def test_hf_group_key_invalid_url(self):
-        """Malformed HF URL returns None."""
-        assert BaseModelService._extract_hf_group_key({"hf_url": "not-a-url"}) is None
-        assert BaseModelService._extract_hf_group_key({"hf_url": "https://example.com"}) is None
-
-    # --- _extract_group_key ---
+class TestSourceGroupKey:
+    """Tests for _extract_group_key (CivitAI id, then site-native source identity)."""
 
     def test_group_key_civitai_only(self):
         """CivitAI modelId returned as int."""
@@ -1303,30 +1272,64 @@ class TestHfGroupKey:
         assert BaseModelService._extract_group_key(item) == 123
 
     def test_group_key_hf_only(self):
-        """HF-only item returns hf:user/repo string."""
+        """HF-linked items never group: a repository is not a model identity."""
         item = {"hf_url": "https://huggingface.co/user/repo"}
-        assert BaseModelService._extract_group_key(item) == "hf:user/repo"
+        assert BaseModelService._extract_group_key(item) is None
 
     def test_group_key_civitai_preferred(self):
-        """CivitAI modelId takes precedence over hf_url."""
+        """CivitAI modelId takes precedence over any source identity."""
         item = {
             "civitai": {"modelId": 456},
-            "hf_url": "https://huggingface.co/other/repo",
+            "source_url": "https://tensor.art/models/789",
         }
         assert BaseModelService._extract_group_key(item) == 456
 
     def test_group_key_neither(self):
-        """No CivitAI or HF returns None."""
+        """No CivitAI or groupable source returns None."""
         assert BaseModelService._extract_group_key({}) is None
         assert BaseModelService._extract_group_key({"some": "data"}) is None
 
     def test_group_key_civitai_none_model_id(self):
-        """civitai.modelId=None falls through to HF."""
+        """civitai.modelId=None falls through to the source identity."""
         item = {
             "civitai": {"modelId": None},
-            "hf_url": "https://huggingface.co/user/repo",
+            "source_url": "https://tensor.art/models/789",
         }
-        assert BaseModelService._extract_group_key(item) == "hf:user/repo"
+        assert BaseModelService._extract_group_key(item) == "ta:789"
+
+    def test_group_key_modelscope_uses_published_model_id(self):
+        """ModelScope groups under ms:<modelId> once enrichment recorded it."""
+        item = {
+            "source_platform": "modelscope",
+            "source_url": "https://modelscope.cn/models/u/r",
+            "source_model_id": "555",
+        }
+        assert BaseModelService._extract_group_key(item) == "ms:555"
+
+    def test_group_key_modelscope_unenriched_stays_standalone(self):
+        """Without source_model_id there is no key — never repo-level grouping."""
+        item = {
+            "source_platform": "modelscope",
+            "source_url": "https://modelscope.cn/models/u/r",
+        }
+        assert BaseModelService._extract_group_key(item) is None
+
+    def test_group_key_modelscope_identity_crosses_repos(self):
+        """Same published-model id groups across repos; same repo does not."""
+
+        def ms_item(repo, model_id):
+            return {
+                "source_platform": "modelscope",
+                "source_url": f"https://modelscope.cn/models/{repo}",
+                "source_model_id": model_id,
+            }
+
+        assert BaseModelService._extract_group_key(
+            ms_item("alice/collection", "555")
+        ) == BaseModelService._extract_group_key(ms_item("bob/mirror", "555"))
+        assert BaseModelService._extract_group_key(
+            ms_item("alice/collection", "555")
+        ) != BaseModelService._extract_group_key(ms_item("alice/collection", "777"))
 
 
 class TestApplyHashFilters:

tests/services/test_model_scanner.py

@@ -967,6 +967,8 @@ def _make_cache_entry(**overrides) -> Dict[str, Any]:
         "source_platform": "",
         "source_url": "",
         "hf_url": "",
+        "source_model_id": "",
+        "source_version_id": "",
         "license_flags": 113,
         "hash_status": "completed",
     }
@@ -1005,6 +1007,8 @@ async def test_sync_cache_no_change(tmp_path: Path):
         "tags": ["alpha"],
         "civitai": {"id": 111, "modelId": 222, "name": "v1"},
         "hf_url": "",
+        "source_model_id": "",
+        "source_version_id": "",
     }
 
     changed = await scanner.sync_cache_from_metadata(
@@ -1049,6 +1053,8 @@ async def test_sync_cache_in_place_update(tmp_path: Path):
         "tags": ["beta", "gamma"],
         "civitai": {"id": 111, "modelId": 222, "name": "v1"},
         "hf_url": "",
+        "source_model_id": "",
+        "source_version_id": "",
     }
 
     changed = await scanner.sync_cache_from_metadata(
@@ -1094,6 +1100,8 @@ async def test_sync_cache_not_in_cache_delegates(tmp_path: Path):
         "tags": [],
         "civitai": {},
         "hf_url": "",
+        "source_model_id": "",
+        "source_version_id": "",
     }
 
     changed = await scanner.sync_cache_from_metadata(
@@ -1147,6 +1155,8 @@ async def test_sync_cache_conditional_resort_skipped(tmp_path: Path, monkeypatch
         "tags": ["alpha"],
         "civitai": {"id": 111, "modelId": 222, "name": "v1"},
         "hf_url": "",
+        "source_model_id": "",
+        "source_version_id": "",
     }
 
     changed = await scanner.sync_cache_from_metadata(
@@ -1197,6 +1207,8 @@ async def test_sync_cache_conditional_resort_triggered(tmp_path: Path, monkeypat
         "tags": ["alpha"],
         "civitai": {"id": 111, "modelId": 222, "name": "v1"},
         "hf_url": "",
+        "source_model_id": "",
+        "source_version_id": "",
     }
 
     changed = await scanner.sync_cache_from_metadata(

tests/services/test_model_sources.py

@@ -279,15 +279,39 @@ class TestHelpers:
         assert get_source_platform({"source_platform": "tensorart"}) == "tensorart"
         assert get_source_platform({}) == ""
 
-    def test_group_keys_match_legacy_hf_shape(self):
-        assert source_group_key({"hf_url": "https://huggingface.co/u/r"}) == "hf:u/r"
-        assert (
-            source_group_key({"source_url": "https://modelscope.cn/models/u/r"}) == "ms:u/r"
-        )
+    def test_group_keys_use_site_native_identity(self):
+        # Hugging Face has no site-native model identity: never grouped.
+        assert source_group_key({"hf_url": "https://huggingface.co/u/r"}) is None
+        # TensorArt's numeric id already identifies a single model.
         assert (
             source_group_key({"source_url": "https://tensor.art/models/123"}) == "ta:123"
         )
 
+    def test_modelscope_groups_by_published_model_id(self):
+        # Without an enriched source_model_id the model stays standalone —
+        # never grouped by repo, which would collapse a collection repo.
+        assert (
+            source_group_key({"source_url": "https://modelscope.cn/models/u/r"}) is None
+        )
+        assert (
+            source_group_key(
+                {
+                    "source_url": "https://modelscope.cn/models/u/r",
+                    "source_model_id": "555",
+                }
+            )
+            == "ms:555"
+        )
+        assert (
+            source_group_key(
+                {
+                    "source_url": "https://www.modelscope.ai/models/u/r",
+                    "source_model_id": "678",
+                }
+            )
+            == "msai:678"
+        )
+
     def test_group_key_is_none_without_source(self):
         assert source_group_key({}) is None
         assert source_group_key({"hf_url": "https://example.com/x"}) is None
@@ -457,7 +481,12 @@ def _modelscope_detail_payload() -> dict:
                 "versions": [
                     {
                         "stats": {"fileList": ["Krea-2-LORA_c1-st8000.safetensors"]},
-                        "modelVersion": {"showName": "c1-st8000", "triggerWords": '[""]'},
+                        "modelVersion": {
+                            "showName": "c1-st8000",
+                            "triggerWords": '[""]',
+                            "id": 1001,
+                            "modelId": 555,
+                        },
                         "coverImages": [
                             {"url": "https://resources.modelscope.cn/cover-images/a.png"}
                         ],
@@ -467,6 +496,8 @@ def _modelscope_detail_payload() -> dict:
                         "modelVersion": {
                             "showName": "c1-st1000",
                             "triggerWords": '["kreaface","kreamodel"]',
+                            "id": 1002,
+                            "modelId": 555,
                         },
                         "coverImages": [
                             {"url": "https://resources.modelscope.cn/cover-images/b.png"},
@@ -533,6 +564,9 @@ class TestFetchModelCardContext:
         # The version label is taken from the file that was matched, not from
         # whichever version happens to come first in the payload.
         assert context.version_name == "c1-st1000"
+        # The site-native identity ids belong to the matched version too.
+        assert context.source_model_id == "555"
+        assert context.source_version_id == "1002"
 
     @pytest.mark.asyncio
     async def test_modelscope_version_label_is_empty_for_an_unknown_file(
@@ -550,6 +584,9 @@ class TestFetchModelCardContext:
         )
 
         assert context.version_name == ""
+        # No version matched, so there is no per-version identity either.
+        assert context.source_model_id == ""
+        assert context.source_version_id == ""
         # The repository-wide fields are still published.
         assert context.model_name == "Krea-2-LORA"
 

tests/services/test_post_processor.py

@@ -817,6 +817,75 @@ class TestSiteProvidedContext:
             "https://huggingface.co/user/repo/resolve/main/images/cat.png"
         ]
 
+    @pytest.mark.asyncio
+    async def test_site_identity_ids_are_persisted(self, processor):
+        """source_model_id/source_version_id reach the sidecar for grouping."""
+        context = ModelCardContext(source_model_id="555", source_version_id="1002")
+
+        with (
+            mock.patch("py.metadata_ops.apply_metadata_updates") as mock_apply,
+            mock.patch("py.metadata_ops.download_preview", return_value=None),
+            mock.patch("py.metadata_ops.refresh_cache"),
+        ):
+            await processor.process(
+                skill_name="enrich_hf_metadata",
+                model_path="/p.safetensors",
+                llm_output=self.LLM_OUTPUT,
+                metadata=dict(self.MODELSCOPE_METADATA),
+                readme_content="",
+                source_context=context,
+            )
+
+        applied = mock_apply.call_args[0][1]
+        assert applied["source_model_id"] == "555"
+        assert applied["source_version_id"] == "1002"
+
+    @pytest.mark.asyncio
+    async def test_site_identity_ids_absent_without_context_values(self, processor):
+        """No identity keys are written when the site did not publish any."""
+        with (
+            mock.patch("py.metadata_ops.apply_metadata_updates") as mock_apply,
+            mock.patch("py.metadata_ops.download_preview", return_value=None),
+            mock.patch("py.metadata_ops.refresh_cache"),
+        ):
+            await processor.process(
+                skill_name="enrich_hf_metadata",
+                model_path="/p.safetensors",
+                llm_output=self.LLM_OUTPUT,
+                metadata=dict(self.MODELSCOPE_METADATA),
+                readme_content="",
+                source_context=ModelCardContext(description="summary only"),
+            )
+
+        applied = mock_apply.call_args[0][1]
+        assert "source_model_id" not in applied
+        assert "source_version_id" not in applied
+
+    @pytest.mark.asyncio
+    async def test_site_identity_ids_skipped_for_a_model_with_no_external_source(
+        self, processor
+    ):
+        """A CivitAI-only model must not pick up source identity ids."""
+        context = ModelCardContext(source_model_id="555", source_version_id="1002")
+
+        with (
+            mock.patch("py.metadata_ops.apply_metadata_updates") as mock_apply,
+            mock.patch("py.metadata_ops.download_preview", return_value=None),
+            mock.patch("py.metadata_ops.refresh_cache"),
+        ):
+            await processor.process(
+                skill_name="enrich_hf_metadata",
+                model_path="/p.safetensors",
+                llm_output=self.LLM_OUTPUT,
+                metadata={"from_civitai": True},
+                readme_content="",
+                source_context=context,
+            )
+
+        applied = mock_apply.call_args[0][1]
+        assert "source_model_id" not in applied
+        assert "source_version_id" not in applied
+
 
 
 # ======================================================================