fix(hosted-vllm): honor ssl verify for transcriptions

litellm/litellm_core_utils/get_litellm_params.py

@@ -1,4 +1,4 @@
-from typing import Optional
+from typing import Optional, Union
 
 from litellm.llms.openai.data_residency import infer_openai_data_residency
 
@@ -107,7 +107,7 @@ def get_litellm_params(
     prompt_id: Optional[str] = None,
     prompt_variables: Optional[dict] = None,
     async_call: Optional[bool] = None,
-    ssl_verify: Optional[bool] = None,
+    ssl_verify: Optional[Union[bool, str]] = None,
     merge_reasoning_content_in_choices: Optional[bool] = None,
     use_litellm_proxy: Optional[bool] = None,
     api_version: Optional[str] = None,

litellm/llms/custom_httpx/llm_http_handler.py

@@ -1298,7 +1298,7 @@ class BaseLLMHTTPHandler:
         )
 
         if client is None or not isinstance(client, HTTPHandler):
-            client = _get_httpx_client()
+            client = _get_httpx_client(params={"ssl_verify": litellm_params.get("ssl_verify", None)})
 
         json_data = data if files is None and isinstance(data, dict) else None
 

litellm/llms/hosted_vllm/transcriptions/transformation.py

@@ -6,6 +6,7 @@ from typing import Optional, Union
 
 import httpx
 
+from litellm.litellm_core_utils.audio_utils.utils import process_audio_file
 from litellm.llms.base_llm.audio_transcription.transformation import (
     AudioTranscriptionRequestData,
 )
@@ -54,12 +55,19 @@ class HostedVLLMAudioTranscriptionConfig(OpenAIWhisperAudioTranscriptionConfig):
         optional_params: dict,
         litellm_params: dict,
     ) -> AudioTranscriptionRequestData:
-        """
-        Transform the audio transcription request
-        """
+        processed_audio = process_audio_file(audio_file)
+        extra_body = optional_params.get("extra_body") or {}
+        data = {
+            "model": model,
+            **{key: value for key, value in optional_params.items() if key != "extra_body"},
+            **extra_body,
+        }
+        files = {
+            "file": (
+                processed_audio.filename,
+                processed_audio.file_content,
+                processed_audio.content_type,
+            )
+        }
 
-        data = {"model": model, "file": audio_file, **optional_params}
-
-        return AudioTranscriptionRequestData(
-            data=data,
-        )
+        return AudioTranscriptionRequestData(data=data, files=files)

litellm/main.py

@@ -7620,7 +7620,9 @@ def transcription(
             max_retries=max_retries,
             litellm_params=litellm_params_dict,
         )
-    elif custom_llm_provider == "openai" or (custom_llm_provider in litellm.openai_compatible_providers):
+    elif custom_llm_provider == "openai" or (
+        custom_llm_provider in litellm.openai_compatible_providers and custom_llm_provider != "hosted_vllm"
+    ):
         api_base = (
             api_base
             or litellm.api_base

litellm/types/utils.py

@@ -3205,6 +3205,7 @@ all_litellm_params = (
         "use_chat_completions_api",
         "prompt_label",
         "shared_session",
+        "ssl_verify",
         "search_tool_name",
         "order",
         "enable_json_schema_validation",

tests/test_litellm/llms/hosted_vllm/transcriptions/test_transformation.py

@@ -0,0 +1,59 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import httpx
+import pytest
+
+import litellm
+
+
+def _transcription_response() -> httpx.Response:
+    return httpx.Response(
+        status_code=200,
+        headers={"content-type": "application/json"},
+        json={"text": "Test transcription"},
+    )
+
+
+def test_transcription_passes_custom_ca_to_sync_http_client() -> None:
+    client = MagicMock()
+    client.post.return_value = _transcription_response()
+
+    with patch(
+        "litellm.llms.custom_httpx.llm_http_handler._get_httpx_client",
+        return_value=client,
+    ) as get_httpx_client:
+        response = litellm.transcription(
+            model="hosted_vllm/whisper-1",
+            file=("audio.wav", b"audio", "audio/wav"),
+            api_base="https://vllm.example.com",
+            ssl_verify="/path/to/ca-cert.crt",
+        )
+
+    assert response.text == "Test transcription"
+    assert get_httpx_client.call_args.kwargs["params"]["ssl_verify"] == "/path/to/ca-cert.crt"
+    request = client.post.call_args.kwargs
+    assert request["data"] == {"model": "whisper-1"}
+    assert request["files"] == {"file": ("audio.wav", b"audio", "audio/wav")}
+
+
+@pytest.mark.asyncio
+async def test_transcription_passes_custom_ca_to_async_http_client() -> None:
+    client = MagicMock()
+    client.post = AsyncMock(return_value=_transcription_response())
+
+    with patch(
+        "litellm.llms.custom_httpx.llm_http_handler.get_async_httpx_client",
+        return_value=client,
+    ) as get_async_httpx_client:
+        response = await litellm.atranscription(
+            model="hosted_vllm/whisper-1",
+            file=("audio.wav", b"audio", "audio/wav"),
+            api_base="https://vllm.example.com",
+            ssl_verify="/path/to/ca-cert.crt",
+        )
+
+    assert response.text == "Test transcription"
+    assert get_async_httpx_client.call_args.kwargs["params"]["ssl_verify"] == "/path/to/ca-cert.crt"
+    request = client.post.call_args.kwargs
+    assert request["data"] == {"model": "whisper-1"}
+    assert request["files"] == {"file": ("audio.wav", b"audio", "audio/wav")}