fix(utils): allow httpx.Timeout for mistral, openai-compatible and httpx-based providers (#47822)

litellm/utils.py

@@ -2846,15 +2846,61 @@ def token_counter(
     )
 
 
+HTTPX_TIMEOUT_SUPPORTED_PROVIDERS: Final[frozenset[str]] = frozenset(
+    {
+        "openai",
+        "azure",
+        "bedrock",
+        "bedrock_mantle",
+        "custom_openai",
+        "text-completion-openai",
+        "openai_like",
+        "chatgpt",
+        "anthropic",
+        "anthropic_text",
+        "mistral",
+        "codestral",
+        "cohere",
+        "cohere_chat",
+        "ollama",
+        "ollama_chat",
+        "openrouter",
+        "vertex_ai",
+        "gemini",
+        "deepseek",
+        "groq",
+        "together_ai",
+        "fireworks_ai",
+        "perplexity",
+        "xai",
+        "zai",
+        "cerebras",
+        "sambanova",
+        "baseten",
+        "cloudflare",
+        "databricks",
+        "hosted_vllm",
+        "minimax",
+        "huggingface",
+    }
+)
+
+
 def supports_httpx_timeout(custom_llm_provider: str) -> bool:
     """
     Helper function to know if a provider implementation supports httpx timeout
     """
-    supported_providers: Final = ["openai", "azure", "bedrock"]
-
-    if custom_llm_provider in supported_providers:
+    if custom_llm_provider in HTTPX_TIMEOUT_SUPPORTED_PROVIDERS:
         return True
 
+    try:
+        from litellm.constants import openai_compatible_providers
+
+        if custom_llm_provider in openai_compatible_providers:
+            return True
+    except ImportError:
+        pass
+
     return False
 
 

tests/unit/test_completion_timeout_resolution.py

@@ -134,7 +134,7 @@ def test_httpx_timeout_coerced_for_provider_without_httpx_timeout_support():
     out = CompletionTimeout.resolve(
         t,
         {},
-        "azure_ai",
+        "replicate",
         global_timeout=None,
         supports_httpx_timeout=supports_httpx_timeout,
         stream_kind=StreamKind.NOT_STREAMED,
@@ -182,7 +182,7 @@ def test_sdk_timeout_coerces_read_timeout_for_providers_without_httpx_support(
         CompletionTimeout.resolve(
             SDKTimeout(connect=2.0, read=read_timeout, write=5.0, pool=7.0),
             {},
-            "azure_ai",
+            "replicate",
             global_timeout=None,
             supports_httpx_timeout=supports_httpx_timeout,
             stream_kind=StreamKind.NOT_STREAMED,
@@ -231,3 +231,22 @@ def test_fake_stream_with_only_a_stream_timeout_keeps_the_global_request_timeout
         )
         == 600.0
     )
+
+
+@pytest.mark.parametrize(
+    "provider",
+    ("mistral", "custom_openai", "anthropic", "cohere", "ollama", "groq", "deepseek"),
+)
+def test_httpx_timeout_preserved_for_extended_providers(provider: str) -> None:
+    t = httpx.Timeout(10.0, connect=1.0)
+    out = CompletionTimeout.resolve(
+        t,
+        {},
+        provider,
+        global_timeout=None,
+        supports_httpx_timeout=supports_httpx_timeout,
+        stream_kind=StreamKind.NOT_STREAMED,
+    )
+    assert out is t
+    assert isinstance(out, httpx.Timeout)
+    assert out.connect == 1.0