chore: enable flash and fix vertex integration (#2003)
This commit is contained in:
@@ -1,5 +1,6 @@
|
||||
GOOGLE_MODEL_TO_CONTEXT_LENGTH = {
|
||||
"gemini-2.5-pro-exp-03-25": 1048576,
|
||||
"gemini-2.5-flash-preview-04-17": 1048576,
|
||||
"gemini-2.0-flash-001": 1048576,
|
||||
"gemini-2.0-pro-exp-02-05": 2097152,
|
||||
"gemini-2.0-flash-lite-preview-02-05": 1048576,
|
||||
|
||||
@@ -110,12 +110,12 @@ class GoogleVertexClient(GoogleAIClient):
|
||||
for candidate in response.candidates:
|
||||
content = candidate.content
|
||||
|
||||
if "role" not in content or not content["role"]:
|
||||
# This means the response is malformed like MALFORMED_FUNCTION_CALL
|
||||
# NOTE: must be a ValueError to trigger a retry
|
||||
raise ValueError(f"Error in response data from LLM: {response_data}")
|
||||
role = content["role"]
|
||||
assert role == "model", f"Unknown role in response: {role}"
|
||||
# if "role" not in content or not content["role"]:
|
||||
# # This means the response is malformed like MALFORMED_FUNCTION_CALL
|
||||
# # NOTE: must be a ValueError to trigger a retry
|
||||
# raise ValueError(f"Error in response data from LLM: {response_data}")
|
||||
# role = content["role"]
|
||||
# assert role == "model", f"Unknown role in response: {role}"
|
||||
|
||||
parts = content.parts
|
||||
|
||||
|
||||
Reference in New Issue
Block a user