feat: download Hugging Face GGUF models

tests/test_validation_harness.py

@@ -121,12 +121,48 @@ class ValidationHarnessTests(unittest.TestCase):
         with patch.object(api, "_search_ollama_catalog", return_value=ollama) as ollama_search, patch.object(
             api, "_search_huggingface", return_value=huggingface
         ) as hf_search:
-            response = api.catalog_search("Qwen3.8", limit=10)
+            response = api.catalog_search("Qwen3.8", limit=10, include_huggingface=True)
         self.assertEqual(response["results"], ollama + huggingface)
         self.assertEqual(response["sources"], ["ollama", "huggingface"])
         ollama_search.assert_called_once_with("Qwen3.8", limit=10)
         hf_search.assert_called_once_with("Qwen3.8", limit=10)
 
+    def test_huggingface_groups_split_gguf_files_into_complete_sets(self):
+        metadata = {
+            "siblings": [
+                {"rfilename": "Q4/Qwen-00001-of-00002.gguf"},
+                {"rfilename": "Q4/Qwen-00002-of-00002.gguf"},
+                {"rfilename": "Q8/Qwen.gguf"},
+                {"rfilename": "Q8/README.md"},
+            ]
+        }
+        with patch.object(api, "_json_request", return_value=metadata):
+            files = api._huggingface_repo_files("owner/repository")
+        self.assertEqual(len(files), 2)
+        split = next(item for item in files if item["split"])
+        self.assertEqual(split["file_count"], 2)
+        self.assertEqual(len(split["filenames"]), 2)
+        single = next(item for item in files if not item["split"])
+        self.assertEqual(single["filename"], "Q8/Qwen.gguf")
+
+    def test_huggingface_download_rejects_non_gguf_paths(self):
+        with self.assertRaises(api.HTTPException):
+            api._valid_huggingface_filename("../model.safetensors")
+        with self.assertRaises(api.HTTPException):
+            api._valid_huggingface_filename("model.bin")
+        self.assertEqual(api._valid_huggingface_filename("Q4_K_M/model.gguf"), "Q4_K_M/model.gguf")
+
+    def test_huggingface_download_queues_validated_file(self):
+        body = api.HuggingFaceDownloadRequest(repo_id="unsloth/Qwen3.8-Flash-Next-GGUF", filename="Q4_K_M/model.gguf")
+        fake_thread = type("Thread", (), {"start": lambda self: None})
+        with patch.object(api, "_huggingface_repo_files", return_value=[{"filename": body.filename, "size": 123}]), patch.object(
+            api.threading, "Thread", return_value=fake_thread()
+        ) as thread:
+            response = api.huggingface_download(body)
+        self.assertTrue(response["ok"])
+        self.assertEqual(response["filename"], body.filename)
+        thread.assert_called_once()
+
     def test_primary_draft_validators_and_primary_compilation_produce_one_answer(self):
         body = api.ChatRequest(
             primary_model="primary",