Merge pull request #39237 from BerriAI/litellm_fix_dashscope_rerank_endpoint

fix(dashscope): remap chat-shaped api_base to the live rerank route

litellm/llms/dashscope/common_utils.py

@@ -2,7 +2,8 @@
 Common utilities for the DashScope LLM provider.
 """
 
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Final
+from urllib.parse import urlparse
 
 import httpx
 
@@ -16,6 +17,27 @@ if TYPE_CHECKING:
     )
     from litellm.llms.base_llm.rerank.transformation import BaseRerankConfig
 
+DASHSCOPE_CHAT_COMPATIBLE_PATH: Final = "/compatible-mode/v1"
+DASHSCOPE_RERANK_PATH: Final = "/compatible-api/v1/reranks"
+
+
+def _rerank_base_for_chat_shaped_base(api_base: str | None) -> str | None:
+    if api_base is None:
+        return None
+    parsed: Final = urlparse(api_base)
+    host: Final = parsed.hostname or ""
+    on_aliyun_host: Final = host == "aliyuncs.com" or host.endswith(".aliyuncs.com")
+    if not on_aliyun_host or parsed.path.rstrip("/") != DASHSCOPE_CHAT_COMPATIBLE_PATH:
+        return None
+    return f"{parsed.scheme}://{parsed.netloc}{DASHSCOPE_RERANK_PATH}"
+
+
+def resolve_dashscope_family_rerank_api_base(api_base: str | None, env_var: str, default_rerank_base: str) -> str:
+    remapped: Final = _rerank_base_for_chat_shaped_base(api_base)
+    if api_base is not None and remapped is None:
+        return api_base
+    return get_secret_str(env_var) or remapped or default_rerank_base
+
 
 def get_dashscope_family_embedding_config(custom_llm_provider: str) -> "BaseEmbeddingConfig":
     if custom_llm_provider == "qwencloud":

litellm/llms/dashscope/qwen_ai_platform.py

@@ -3,6 +3,7 @@ from typing import Final
 from litellm.secret_managers.main import get_secret_str
 
 from .chat.transformation import DashScopeChatConfig
+from .common_utils import resolve_dashscope_family_rerank_api_base
 from .embed.transformation import DashScopeEmbeddingConfig
 from .image_generation.transformation import DashScopeImageGenerationConfig
 from .rerank.transformation import DashScopeRerankConfig
@@ -51,7 +52,9 @@ class QwenAIPlatformRerankConfig(DashScopeRerankConfig):
         return _require_qwen_ai_platform_api_key(api_key)
 
     def _resolve_rerank_api_base(self, api_base: str | None) -> str:
-        return api_base or get_secret_str("QWEN_AI_PLATFORM_API_BASE_RERANK") or QWEN_AI_PLATFORM_RERANK_API_BASE
+        return resolve_dashscope_family_rerank_api_base(
+            api_base, "QWEN_AI_PLATFORM_API_BASE_RERANK", QWEN_AI_PLATFORM_RERANK_API_BASE
+        )
 
 
 class QwenAIPlatformImageGenerationConfig(DashScopeImageGenerationConfig):

litellm/llms/dashscope/qwencloud.py

@@ -3,6 +3,7 @@ from typing import Final
 from litellm.secret_managers.main import get_secret_str
 
 from .chat.transformation import DashScopeChatConfig
+from .common_utils import resolve_dashscope_family_rerank_api_base
 from .embed.transformation import DashScopeEmbeddingConfig
 from .image_generation.transformation import DashScopeImageGenerationConfig
 from .rerank.transformation import DashScopeRerankConfig
@@ -51,7 +52,9 @@ class QwenCloudRerankConfig(DashScopeRerankConfig):
         return _require_qwencloud_api_key(api_key)
 
     def _resolve_rerank_api_base(self, api_base: str | None) -> str:
-        return api_base or get_secret_str("QWENCLOUD_API_BASE_RERANK") or QWENCLOUD_RERANK_API_BASE
+        return resolve_dashscope_family_rerank_api_base(
+            api_base, "QWENCLOUD_API_BASE_RERANK", QWENCLOUD_RERANK_API_BASE
+        )
 
 
 class QwenCloudImageGenerationConfig(DashScopeImageGenerationConfig):

litellm/llms/dashscope/rerank/transformation.py

@@ -12,8 +12,12 @@ Endpoint
 - https://dashscope.aliyuncs.com/compatible-api/v1/reranks
 
 Note: chat/embed live under `/compatible-mode/v1/`, but DashScope's rerank
-route is exposed under `/compatible-api/v1/reranks` per the docs. Override
-with `DASHSCOPE_API_BASE_RERANK` to point at a different host or path.
+route is exposed under `/compatible-api/v1/reranks` per the docs. A chat-shaped
+`.aliyuncs.com/compatible-mode/v1` base reaching this config (the chat default
+from `get_llm_provider`, or a `DASHSCOPE_API_BASE` env var) is redirected to
+the same host's rerank route, since `/compatible-mode/v1/reranks` is a dead
+route on every DashScope host. Override with `DASHSCOPE_API_BASE_RERANK` to
+point at a different host or path.
 
 Empirically, qwen3-rerank accepts `return_documents=true` and echoes
 `results[].document.text` back, even though the public docs list the flag
@@ -40,7 +44,7 @@ from litellm.types.rerank import (
     RerankTokens,
 )
 
-from ..common_utils import DashScopeError
+from ..common_utils import DashScopeError, resolve_dashscope_family_rerank_api_base
 
 DEFAULT_RERANK_URL: Final = "https://dashscope.aliyuncs.com/compatible-api/v1/reranks"
 
@@ -67,9 +71,7 @@ class DashScopeRerankConfig(BaseRerankConfig):
         return resolved_api_key
 
     def _resolve_rerank_api_base(self, api_base: str | None) -> str:
-        if api_base is not None:
-            return api_base
-        return get_secret_str("DASHSCOPE_API_BASE_RERANK") or DEFAULT_RERANK_URL
+        return resolve_dashscope_family_rerank_api_base(api_base, "DASHSCOPE_API_BASE_RERANK", DEFAULT_RERANK_URL)
 
     def get_complete_url(
         self,

tests/test_litellm/llms/dashscope/test_dashscope_rerank_transformation.py

@@ -25,19 +25,43 @@ class TestDashScopeRerankURL:
         url = self.config.get_complete_url(api_base=None, model="qwen3-rerank")
         assert url == DEFAULT_RERANK_URL
 
-    def test_explicit_v1_base_appends_reranks(self):
+    def test_chat_shaped_base_remaps_to_rerank_route(self):
         url = self.config.get_complete_url(
             api_base="https://dashscope.aliyuncs.com/compatible-mode/v1",
             model="qwen3-rerank",
         )
-        assert url == "https://dashscope.aliyuncs.com/compatible-mode/v1/reranks"
+        assert url == "https://dashscope.aliyuncs.com/compatible-api/v1/reranks"
 
-    def test_intl_v1_base_appends_reranks(self):
+    def test_intl_chat_shaped_base_remaps_to_intl_rerank_route(self):
         url = self.config.get_complete_url(
             api_base="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
             model="qwen3-rerank",
         )
-        assert url == "https://dashscope-intl.aliyuncs.com/compatible-mode/v1/reranks"
+        assert url == "https://dashscope-intl.aliyuncs.com/compatible-api/v1/reranks"
+
+    def test_chat_shaped_base_with_trailing_slash_remaps(self):
+        url = self.config.get_complete_url(
+            api_base="https://dashscope-intl.aliyuncs.com/compatible-mode/v1/",
+            model="qwen3-rerank",
+        )
+        assert url == "https://dashscope-intl.aliyuncs.com/compatible-api/v1/reranks"
+
+    def test_rerank_env_var_wins_over_chat_shaped_base(self, monkeypatch):
+        monkeypatch.setenv(
+            "DASHSCOPE_API_BASE_RERANK", "https://rerank.example.com/v1/reranks"
+        )
+        url = self.config.get_complete_url(
+            api_base="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
+            model="qwen3-rerank",
+        )
+        assert url == "https://rerank.example.com/v1/reranks"
+
+    def test_non_aliyun_chat_path_base_not_remapped(self):
+        url = self.config.get_complete_url(
+            api_base="https://gateway.example.com/compatible-mode/v1",
+            model="qwen3-rerank",
+        )
+        assert url == "https://gateway.example.com/compatible-mode/v1/reranks"
 
     def test_already_complete_url_passthrough(self):
         full = "https://dashscope.aliyuncs.com/compatible-api/v1/reranks"

tests/test_litellm/llms/dashscope/test_qwen_brand_aliases.py

@@ -209,6 +209,11 @@ class TestQwenBrandDefaultUrls:
         url = brand["rerank_config"]().get_complete_url(api_base=None, model="gte-rerank-v2")
         assert url == "https://rerank.example.com/v1/reranks"
 
+    @pytest.mark.parametrize("brand", BRAND_CASES)
+    def test_rerank_remaps_chat_shaped_default_base(self, brand):
+        url = brand["rerank_config"]().get_complete_url(api_base=brand["default_base"], model="gte-rerank-v2")
+        assert url == brand["default_rerank_base"]
+
     @pytest.mark.parametrize("brand", BRAND_CASES)
     def test_image_generation_complete_url(self, brand):
         url = brand["image_config"]().get_complete_url(

tests/test_litellm/rerank_api/test_main.py

@@ -111,6 +111,72 @@ def test_together_rerank_honors_api_base(respx_mock: respx.MockRouter):
     assert mock_route.calls[0].request.headers["authorization"] == "Bearer fake-together-key"
 
 
+DASHSCOPE_RERANK_BODY = {
+    "object": "list",
+    "results": [{"index": 0, "relevance_score": 0.95}],
+    "model": "qwen3-rerank",
+    "id": "rerank-mock-id",
+    "usage": {"total_tokens": 10},
+}
+
+
+def test_dashscope_rerank_defaults_to_live_rerank_route(respx_mock: respx.MockRouter, monkeypatch):
+    """Regression for the dead default endpoint: get_llm_provider always returns the
+    chat base for dashscope, which used to hijack rerank onto the dead
+    /compatible-mode/v1/reranks route."""
+    monkeypatch.delenv("DASHSCOPE_API_BASE", raising=False)
+    monkeypatch.delenv("DASHSCOPE_API_BASE_RERANK", raising=False)
+
+    mock_route = respx_mock.post("https://dashscope.aliyuncs.com/compatible-api/v1/reranks")
+    mock_route.return_value = httpx.Response(200, json=DASHSCOPE_RERANK_BODY)
+
+    response = litellm.rerank(
+        model="dashscope/qwen3-rerank",
+        query=MARKER_QUERY,
+        documents=[MARKER_DOC],
+        api_key="fake-dashscope-key",
+    )
+
+    assert mock_route.called
+    assert response.results[0]["relevance_score"] == 0.95
+
+
+def test_dashscope_rerank_chat_env_base_keeps_host_and_rerank_route(respx_mock: respx.MockRouter, monkeypatch):
+    """Regression: a chat-style DASHSCOPE_API_BASE must not hijack rerank onto the
+    chat path, while its host (the region) is preserved."""
+    monkeypatch.setenv("DASHSCOPE_API_BASE", "https://dashscope-intl.aliyuncs.com/compatible-mode/v1")
+    monkeypatch.delenv("DASHSCOPE_API_BASE_RERANK", raising=False)
+
+    mock_route = respx_mock.post("https://dashscope-intl.aliyuncs.com/compatible-api/v1/reranks")
+    mock_route.return_value = httpx.Response(200, json=DASHSCOPE_RERANK_BODY)
+
+    litellm.rerank(
+        model="dashscope/qwen3-rerank",
+        query=MARKER_QUERY,
+        documents=[MARKER_DOC],
+        api_key="fake-dashscope-key",
+    )
+
+    assert mock_route.called
+
+
+def test_dashscope_rerank_explicit_api_base_wins(respx_mock: respx.MockRouter, monkeypatch):
+    monkeypatch.setenv("DASHSCOPE_API_BASE", "https://dashscope-intl.aliyuncs.com/compatible-mode/v1")
+
+    mock_route = respx_mock.post("https://custom-rerank.example/v1/reranks")
+    mock_route.return_value = httpx.Response(200, json=DASHSCOPE_RERANK_BODY)
+
+    litellm.rerank(
+        model="dashscope/qwen3-rerank",
+        query=MARKER_QUERY,
+        documents=[MARKER_DOC],
+        api_key="fake-dashscope-key",
+        api_base="https://custom-rerank.example/v1",
+    )
+
+    assert mock_route.called
+
+
 DASHSCOPE_404_BODY = {
     "error": {
         "message": "The model `does-not-exist` does not exist or you do not have access to it.",