feat: add llama-swap as a backend

test/test_unit_helpers.py

@@ -277,3 +277,49 @@ class TestGetTrackingModel:
         with patch.object(router, "config", cfg):
             result = router.get_tracking_model(ep, "unsloth/model:Q8_0")
         assert result == "model"
+
+
+class TestLlamaSwapClassification:
+    def _cfg(self, *, server=None, swap=None):
+        cfg = MagicMock()
+        cfg.endpoints = []
+        cfg.llama_server_endpoints = server or []
+        cfg.llama_swap_endpoints = swap or []
+        return cfg
+
+    def test_is_llama_swap_only_for_swap_list(self):
+        from backends.normalize import is_llama_swap
+        swap_ep = "http://host:8890/v1"
+        server_ep = "http://host:8889/v1"
+        cfg = self._cfg(server=[server_ep], swap=[swap_ep])
+        with patch.object(router, "config", cfg):
+            assert is_llama_swap(swap_ep) is True
+            assert is_llama_swap(server_ep) is False
+
+    def test_is_llama_server_covers_both(self):
+        from backends.normalize import is_llama_server
+        swap_ep = "http://host:8890/v1"
+        server_ep = "http://host:8889/v1"
+        cfg = self._cfg(server=[server_ep], swap=[swap_ep])
+        with patch.object(router, "config", cfg):
+            assert is_llama_server(swap_ep) is True
+            assert is_llama_server(server_ep) is True
+            assert is_llama_server("http://host:11434") is False
+
+    def test_swap_is_openai_compatible_not_ext(self):
+        swap_ep = "http://host:8890/v1"
+        cfg = self._cfg(swap=[swap_ep])
+        with patch.object(router, "config", cfg):
+            assert router.is_openai_compatible(swap_ep) is True
+            assert router.is_ext_openai_endpoint(swap_ep) is False
+
+    def test_swap_tracking_model_normalized(self):
+        swap_ep = "http://host:8890/v1"
+        cfg = self._cfg(swap=[swap_ep])
+        with patch.object(router, "config", cfg):
+            assert router.get_tracking_model(swap_ep, "unsloth/model:Q8_0") == "model"
+
+    def test_llama_endpoints_dedupes_and_orders(self):
+        from backends.normalize import llama_endpoints
+        cfg = self._cfg(server=["a", "b"], swap=["b", "c"])
+        assert llama_endpoints(cfg) == ["a", "b", "c"]