From 0bd4d323da3a5969a7a3940faef03ecd239d2a98 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Tue, 25 Aug 2026 10:33:40 -0700
Subject: [PATCH] fix(router): resolve provider from api_base in deployment
 validation and acompletion

Router._add_deployment called get_llm_provider without the deployment's api_base, so a config entry with a bare model plus a known OpenAI-compatible endpoint failed startup validation with LLM Provider NOT provided and the proxy returned 400 no healthy deployments for that model group. acompletion had the same gap at request time: it forwarded only base_url into its get_llm_provider call, dropping the api_base kwarg the router passes. Both now forward api_base so endpoint matching resolves the provider the same way sync completion already does
---
 litellm/main.py                   |  2 +-
 litellm/router.py                 |  1 +
 tests/test_litellm/test_main.py   | 13 +++++++
 tests/test_litellm/test_router.py | 62 +++++++++++++++++++++++++++++++
 4 files changed, 77 insertions(+), 1 deletion(-)

diff --git a/litellm/main.py b/litellm/main.py
index d3967473f99..6dfd8c2d675 100644
--- a/litellm/main.py
+++ b/litellm/main.py
@@ -602,7 +602,7 @@ async def acompletion(
         _, custom_llm_provider, _, _ = get_llm_provider(
             model=model,
             custom_llm_provider=custom_llm_provider,
-            api_base=base_url,
+            api_base=kwargs.get("api_base") or base_url,
         )
 
     fallbacks = fallbacks or litellm.model_fallbacks
diff --git a/litellm/router.py b/litellm/router.py
index 6ee474730c9..33da39e2677 100644
--- a/litellm/router.py
+++ b/litellm/router.py
@@ -8341,6 +8341,7 @@ class Router:
             ) = litellm.get_llm_provider(
                 model=deployment.litellm_params.model,
                 custom_llm_provider=deployment.litellm_params.get("custom_llm_provider", None),
+                api_base=deployment.litellm_params.api_base,
             )
             # done reading model["litellm_params"]
             # Check if provider is supported: either in enum or JSON-configured
diff --git a/tests/test_litellm/test_main.py b/tests/test_litellm/test_main.py
index 99b1cc826aa..8f2b06be4b3 100644
--- a/tests/test_litellm/test_main.py
+++ b/tests/test_litellm/test_main.py
@@ -2944,3 +2944,16 @@ def test_a_stream_that_reported_no_usage_is_still_billed(local_cost_map):
     assert cost == pytest.approx(
         _priced_at(rebuilt.usage.prompt_tokens, rebuilt.usage.completion_tokens)
     )
+
+
+@pytest.mark.asyncio
+async def test_acompletion_resolves_provider_from_api_base():
+    response = await litellm.acompletion(
+        model="deepseek-chat",
+        api_base="https://api.deepseek.com/v1",
+        api_key="fake-key",
+        messages=[{"role": "user", "content": "hi"}],
+        mock_response="resolved",
+    )
+
+    assert response.choices[0].message.content == "resolved"
diff --git a/tests/test_litellm/test_router.py b/tests/test_litellm/test_router.py
index 910b874c2ac..df6754cd7ca 100644
--- a/tests/test_litellm/test_router.py
+++ b/tests/test_litellm/test_router.py
@@ -8921,3 +8921,65 @@ class TestAzureBaseModelFallbackLogging:
             deployment=None, received_model_name="my-group", id="azure-base-model-test-id"
         )
         assert model_info["max_input_tokens"] == litellm.model_cost["azure/gpt-4o-mini"]["max_input_tokens"]
+
+
+class TestAddDeploymentApiBaseProviderResolution:
+    def test_bare_model_with_known_api_base_initializes(self):
+        router = litellm.Router(
+            model_list=[
+                {
+                    "model_name": "groq-pinned",
+                    "litellm_params": {
+                        "model": "llama-3.3-70b-versatile",
+                        "api_base": "https://api.groq.com/openai/v1",
+                        "api_key": "fake-key",
+                    },
+                },
+                {
+                    "model_name": "deepseek-pinned",
+                    "litellm_params": {
+                        "model": "deepseek-chat",
+                        "api_base": "https://api.deepseek.com/v1",
+                        "api_key": "fake-key",
+                    },
+                },
+            ]
+        )
+
+        model_list = router.get_model_list()
+        assert model_list is not None
+        assert {m["model_name"] for m in model_list} == {"groq-pinned", "deepseek-pinned"}
+
+    def test_bare_model_with_unknown_api_base_still_raises(self):
+        with pytest.raises(litellm.BadRequestError, match="LLM Provider NOT provided"):
+            litellm.Router(
+                model_list=[
+                    {
+                        "model_name": "mystery",
+                        "litellm_params": {
+                            "model": "some-unknown-model",
+                            "api_base": "https://llm.internal.example.com/v1",
+                            "api_key": "fake-key",
+                        },
+                    }
+                ]
+            )
+
+    def test_explicit_custom_llm_provider_beats_api_base_endpoint_match(self):
+        router = litellm.Router(
+            model_list=[
+                {
+                    "model_name": "openai-via-gateway",
+                    "litellm_params": {
+                        "model": "gpt-3.5-turbo",
+                        "custom_llm_provider": "openai",
+                        "api_base": "https://api.groq.com/openai/v1",
+                        "api_key": "fake-key",
+                    },
+                }
+            ]
+        )
+
+        deployment = router.get_deployment_by_model_group_name("openai-via-gateway")
+        assert deployment is not None
+        assert deployment.litellm_params.custom_llm_provider == "openai"