[OPIK-2068] [P SDK] LangGraph Studio Failed to extract token usage from presumably Google LLM langchain run. (#2798)
frostbyte_neo
发布于
2025-07-23 13:04:02 +00:00
- [OPIK-2068]: Refactored token usage logic and added LangChain integration support.
- Extracted common token flattening and filtering logic into
flatten_result_and_add_model_extra for reuse across providers.
- Added the
LangChainUsage model for handling LangChain responses, supporting token details and backend-compatible dictionary conversion.
- Integrated
LangChainUsage into the OpikUsage factory, enabling seamless support for LangChain responses in the usage pipeline.
- [OPIK-2068]: Added unit tests for
LangChainUsage.
- Introduced comprehensive test coverage for
LangChainUsage model.
- Validated happy paths, missing input/output token details, and invalid data scenarios.
- Added tests for backend-compatible flat dictionary conversion.
- [OPIK-2068]: Refactored LangChain integration and enhanced token usage extraction.
- Added
_assert_chat_usage_validity and updated tests to validate additional token usage keys.
- Introduced
langchain_run_helpers for modular handling of LangChain-specific token usage and metadata extraction.
- Refactored token usage logic across integrations to leverage
langchain_run_helpers.
- Fixed
completion_tokens handling in OpikUsage to address token calculation inaccuracies.
- Improved maintainability by reducing code duplication and centralizing LangChain token extraction logic.
- [OPIK-2068]: Added Google GenAI integration tests and updated LangChain configuration.
- Introduced
ensure_google_api_configured fixture to verify GOOGLE_API_KEY environment variable setup.
- Added tests validating LangChain's Google GenAI token usage and integration.
- Updated GitHub Actions to include
GOOGLE_API_KEY as a secret.
- Added
langchain_google_genai to library integration requirements.
- [OPIK-2068]: Updated library requirements to fix the Python 3.9 type annotation evaluation issue.
- Added
eval-type-backport to openai integration test requirements for compatibility with Python 3.9.
- [OPIK-2068]: Updated
gemini model version in sample agent SSE.
- Changed model from
gemini-2.5-flash-preview-05-20 to gemini-2.5-flash in weather_time_agent.
- [OPIK-2068]: Trimmed leading and trailing spaces from feedback score reasons in E2E tests.
- Applied
strip() to score.reason in verifiers.py and test_threads_evaluate.py to ensure consistent formatting.
- [OPIK-2068]: Removed
LangChainUsage and refactored token usage handling across integrations.
- Deleted
LangChainUsage model and related tests.
- Refactored
OpikUsage to directly handle LangChain token usage via Google Gemini mapping.
- Consolidated token usage logic into integration-specific helpers for consistent extraction.
- Added
google_helpers.py for reusable token usage validation utilities in tests.
- [OPIK-2068]: Removed
LangChainUsage and refactored token usage handling across integrations.
- Deleted
LangChainUsage model and related tests.
- Refactored
OpikUsage to directly handle LangChain token usage via Google Gemini mapping.
- Consolidated token usage logic into integration-specific helpers for consistent extraction.
- Added
google_helpers.py for reusable token usage validation utilities in tests.
下载附件