scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-09 22:31:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 87

fix(vertex_ai/realtime): drop follow-up session.update to avoid 1007 close

Browse source 
Vertex AI Live treats setup as a first-and-only client message; emitting a
second setup with realtimeInputConfig only closes the websocket with a 1007
policy error. Reverting the follow-up-setup branch restores the pre-existing
no-op behavior for subsequent session.update messages.
This commit is contained in:
mateo-berri 2026-05-22 18:10:56 +00:00
parent 159982a892
commit bcba880c8b
No known key found for this signature in database
1 changed files with 6 additions and 32 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

38
litellm/llms/vertex_ai/realtime/transformation.py
View file

@ -12,15 +12,10 @@ Auth: OAuth2 Bearer token (not an API key).
"""
import json
from typing import List, Optional, cast
from typing import List, Optional
from litellm import verbose_logger
from litellm.llms.gemini.realtime.transformation import GeminiRealtimeConfig
from litellm.types.llms.gemini import (
BidiGenerateContentRealtimeInputConfig,
BidiGenerateContentSetup,
)
from litellm.types.llms.openai import OpenAIRealtimeTurnDetection
class VertexAIRealtimeConfig(GeminiRealtimeConfig):
@ -209,15 +204,17 @@ class VertexAIRealtimeConfig(GeminiRealtimeConfig):
"""
Translate OpenAI realtime client messages to Vertex AI format.
Handles session.update by sending setup with proper Vertex AI model path.
On the first ``session.update`` (when no setup has been sent yet) the
full ``BidiGenerateContentSetup`` is built with Vertex AI's model path
and forwarded. Any later ``session.update`` is dropped: Vertex AI
documents ``setup`` as the first-and-only client message, and a second
``setup`` closes the connection with a 1007 policy error.
"""
json_message = json.loads(message)
msg_type = json_message.get("type")
# Handle session.update with Vertex AI specific model path
if msg_type == "session.update":
if session_configuration_request is None:
# First session.update - send the setup with Vertex AI configuration
setup_config = self._build_vertex_ai_setup_config(
model, json_message.get("session") or {}
)
@ -228,34 +225,11 @@ class VertexAIRealtimeConfig(GeminiRealtimeConfig):
)
return [gemini_setup_msg]
# Subsequent session.update: forward turn_detection-only updates
# (e.g. guardrail-injected disable of VAD auto-response) as a
# follow-up setup. Other fields are dropped because Vertex AI
# doesn't support dynamic updates.
session_payload = json_message.get("session") or {}
turn_detection = self._extract_turn_detection(session_payload)
if turn_detection is not None:
transformed_audio_activity_config = self.map_automatic_turn_detection(
cast(OpenAIRealtimeTurnDetection, turn_detection)
)
if len(transformed_audio_activity_config) > 0:
follow_up_setup: BidiGenerateContentSetup = {
"model": self._vertex_model_path(model),
"realtimeInputConfig": BidiGenerateContentRealtimeInputConfig(
automaticActivityDetection=transformed_audio_activity_config
),
}
verbose_logger.debug(
"Vertex AI Realtime: Forwarding turn_detection-only session.update as setup"
)
return [json.dumps({"setup": follow_up_setup})]
verbose_logger.debug(
"Vertex AI Realtime: Ignoring session.update (setup already sent)"
)
return []
# For other message types, use parent's logic
return super().transform_realtime_request(
message, model, session_configuration_request
)