fix(embeddings): honor the caller's timeout on Cohere sync and HuggingFace embedding requests

Bedrock Cohere embeddings ran the sync request with the client default instead of the timeout
passed to litellm.embedding, and HuggingFace embeddings dropped it on both the sync and async paths.
Pass the timeout through to the HTTP request on each path.

Fixes #47301
Fixes #47299

litellm/llms/cohere/embed/handler.py

@@ -82,7 +82,7 @@ async def async_embedding(
         )
 
     try:
-        response: Final = await client.post(api_base, headers=headers, data=json.dumps(data))
+        response: Final = await client.post(api_base, headers=headers, data=json.dumps(data), timeout=timeout)
     except httpx.HTTPStatusError as e:
         ## LOGGING
         logging_obj.post_call(
@@ -166,7 +166,7 @@ def embedding(
     if client is None or not isinstance(client, HTTPHandler):
         client = HTTPHandler(concurrent_limit=1)
 
-    response: Final = client.post(embed_url, headers=headers, data=json.dumps(data))
+    response: Final = client.post(embed_url, headers=headers, data=json.dumps(data), timeout=timeout)
 
     return CohereEmbeddingConfig().transform_response(
         response=response,

litellm/llms/huggingface/embedding/handler.py

@@ -269,9 +269,10 @@ class HuggingFaceEmbedding(BaseLLM):
         if client is None:
             client = get_async_httpx_client(
                 llm_provider=litellm.LlmProviders.HUGGINGFACE,
+                params={"timeout": timeout},
             )
 
-        response: Final = await client.post(api_base, headers=headers, data=json.dumps(data))
+        response: Final = await client.post(api_base, headers=headers, data=json.dumps(data), timeout=timeout)
 
         ## LOGGING
         logging_obj.post_call(
@@ -374,7 +375,7 @@ class HuggingFaceEmbedding(BaseLLM):
         ## COMPLETION CALL
         if client is None or not isinstance(client, HTTPHandler):
             client = HTTPHandler(concurrent_limit=1)
-        response: Final = client.post(embed_url, headers=headers, data=json.dumps(data))
+        response: Final = client.post(embed_url, headers=headers, data=json.dumps(data), timeout=timeout)
 
         ## LOGGING
         logging_obj.post_call(

litellm/main.py

@@ -7118,6 +7118,7 @@ def embedding(
                 aembedding=aembedding,
                 litellm_params=litellm_params_dict,
                 headers=headers,
+                timeout=timeout,
             )
         elif custom_llm_provider == "bedrock":
             if isinstance(input, str):

tests/unit/llms/cohere/embed/test_cohere_embedding_handler.py

@@ -0,0 +1,35 @@
+from typing import Final
+
+import httpx
+import pytest
+import respx
+
+import litellm
+
+BEDROCK_COHERE_INVOKE_URL: Final = (
+    "https://bedrock-runtime.us-east-1.amazonaws.com/model/cohere.embed-english-v3/invoke"
+)
+COHERE_EMBED_RESPONSE: Final = {"embeddings": [[0.1, 0.2]], "meta": {"billed_units": {"input_tokens": 1}}}
+
+
+@pytest.mark.parametrize("sync_mode", [True, False])
+async def test_bedrock_cohere_embedding_sends_the_caller_timeout(
+    respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, sync_mode: bool
+) -> None:
+    monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
+    route: Final = respx_mock.post(BEDROCK_COHERE_INVOKE_URL).mock(
+        return_value=httpx.Response(200, json=COHERE_EMBED_RESPONSE)
+    )
+    request: Final = {
+        "model": "bedrock/cohere.embed-english-v3",
+        "input": ["hello"],
+        "timeout": 45,
+        "aws_region_name": "us-east-1",
+        "aws_access_key_id": "mockaws_access_key_id",
+        "aws_secret_access_key": "mockaws_secret_access_key",
+    }
+
+    response: Final = litellm.embedding(**request) if sync_mode else await litellm.aembedding(**request)
+
+    assert route.calls.last.request.extensions["timeout"]["read"] == 45
+    assert response.data[0]["embedding"] == [0.1, 0.2]

tests/unit/llms/huggingface/embedding/test_huggingface_embedding_handler.py

@@ -1,12 +1,12 @@
 import json
 from typing import Final
-from unittest.mock import patch, MagicMock, AsyncMock
+from unittest.mock import AsyncMock, MagicMock, patch
 
-
-import litellm
+import httpx
 import pytest
 import respx
 
+import litellm
 
 MOCK_EMBEDDING_RESPONSE = [[0.1, 0.2, 0.3, 0.4, 0.5]]
 
@@ -197,3 +197,30 @@ async def test_hf_embedding_sentence_sim_sends_one_similarity_request(
     }
     assert inference.calls.last.request.headers["Authorization"] == "Bearer hf-test-token"
     assert response.data == [{"object": "embedding", "index": 0, "embedding": 0.7708950042724609}]
+
+
+HF_EMBED_URL: Final = "https://router.huggingface.co/hf-inference/pipeline/embed/BAAI/bge-m3"
+
+
+@pytest.mark.respx(assert_all_called=False)
+@pytest.mark.parametrize("sync_mode", [True, False])
+async def test_huggingface_embedding_sends_the_caller_timeout(
+    respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, sync_mode: bool
+) -> None:
+    monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
+    respx_mock.get(url__regex=r"https://huggingface\.co/.*/config\.json").respond(json={"max_position_embeddings": 512})
+    route: Final = respx_mock.post(HF_EMBED_URL).mock(
+        return_value=httpx.Response(200, json=MOCK_EMBEDDING_RESPONSE)
+    )
+    request: Final = {
+        "model": "huggingface/BAAI/bge-m3",
+        "input": ["hello"],
+        "input_type": "embed",
+        "api_key": "hf_test",
+        "timeout": 45,
+    }
+
+    response: Final = litellm.embedding(**request) if sync_mode else await litellm.aembedding(**request)
+
+    assert route.calls.last.request.extensions["timeout"]["read"] == 45
+    assert response.data[0]["embedding"] == MOCK_EMBEDDING_RESPONSE[0]