scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-09 03:18:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 109

fix: pass credentials to PredictionServiceClient for Vertex AI custom endpoints (#17757)

Browse source 
When calling Vertex AI Model Garden custom endpoints with service account
credentials (instead of ADC), the credentials were not passed to the
PredictionServiceClient, resulting in "default credentials not found" error.

- Move credential loading before cache check to ensure availability
- Pass credentials to sync/async PredictionServiceClient
- Add vertex_credentials param to async_completion and async_streaming

Closes #8597
This commit is contained in:
dongbin-lunark 2025-12-14 12:11:43 +09:00 • committed by GitHub
parent c892c2c83d
commit 0be6dd57c3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 26 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

44
litellm/llms/vertex_ai/vertex_ai_non_gemini.py
View file

@ -137,22 +137,24 @@ def completion( # noqa: PLR0915
)
_vertex_llm_model_object = _get_client_from_cache(client_cache_key=_cache_key)
if _vertex_llm_model_object is None:
from google.auth.credentials import Credentials
# Load credentials - needed for both vertexai.init() and PredictionServiceClient
from google.auth.credentials import Credentials
if vertex_credentials is not None and isinstance(vertex_credentials, str):
import google.oauth2.service_account
if vertex_credentials is not None and isinstance(vertex_credentials, str):
import google.oauth2.service_account
json_obj = json.loads(vertex_credentials)
json_obj = json.loads(vertex_credentials)
creds = (
google.oauth2.service_account.Credentials.from_service_account_info(
json_obj,
scopes=["https://www.googleapis.com/auth/cloud-platform"],
)
creds = (
google.oauth2.service_account.Credentials.from_service_account_info(
json_obj,
scopes=["https://www.googleapis.com/auth/cloud-platform"],
)
else:
creds, _ = google.auth.default(quota_project_id=vertex_project)
)
else:
creds, _ = google.auth.default(quota_project_id=vertex_project)
if _vertex_llm_model_object is None:
print_verbose(
f"VERTEX AI: creds={creds}; google application credentials: {os.getenv('GOOGLE_APPLICATION_CREDENTIALS')}"
)
@ -268,6 +270,7 @@ def completion( # noqa: PLR0915
"instances": instances,
"vertex_location": vertex_location,
"vertex_project": vertex_project,
"vertex_credentials": creds,
"safety_settings": safety_settings,
**optional_params,
}
@ -371,9 +374,10 @@ def completion( # noqa: PLR0915
},
)
llm_model = aiplatform.gapic.PredictionServiceClient(
client_options=client_options
client_options=client_options,
credentials=creds,
)
request_str += f"llm_model = aiplatform.gapic.PredictionServiceClient(client_options={client_options})\n"
request_str += f"llm_model = aiplatform.gapic.PredictionServiceClient(client_options={client_options}, credentials=...)\n"
endpoint_path = llm_model.endpoint_path(
project=vertex_project, location=vertex_location, endpoint=model
)
@ -498,6 +502,7 @@ async def async_completion( # noqa: PLR0915
instances=None,
vertex_project=None,
vertex_location=None,
vertex_credentials=None,
safety_settings=None,
**optional_params,
):
@ -557,9 +562,10 @@ async def async_completion( # noqa: PLR0915
)
llm_model = aiplatform.gapic.PredictionServiceAsyncClient(
client_options=client_options
client_options=client_options,
credentials=vertex_credentials,
)
request_str += f"llm_model = aiplatform.gapic.PredictionServiceAsyncClient(client_options={client_options})\n"
request_str += f"llm_model = aiplatform.gapic.PredictionServiceAsyncClient(client_options={client_options}, credentials=...)\n"
endpoint_path = llm_model.endpoint_path(
project=vertex_project, location=vertex_location, endpoint=model
)
@ -661,6 +667,7 @@ async def async_streaming( # noqa: PLR0915
instances=None,
vertex_project=None,
vertex_location=None,
vertex_credentials=None,
safety_settings=None,
**optional_params,
):
@ -724,9 +731,10 @@ async def async_streaming( # noqa: PLR0915
},
)
llm_model = aiplatform.gapic.PredictionServiceAsyncClient(
client_options=client_options
client_options=client_options,
credentials=vertex_credentials,
)
request_str += f"llm_model = aiplatform.gapic.PredictionServiceAsyncClient(client_options={client_options})\n"
request_str += f"llm_model = aiplatform.gapic.PredictionServiceAsyncClient(client_options={client_options}, credentials=...)\n"
endpoint_path = llm_model.endpoint_path(
project=vertex_project, location=vertex_location, endpoint=model
)