The langchain-fireworks integration has been updated to version 1.6.0, bringing new capabilities and important fixes for self-hosted LangChain deployments. The headline feature is the addition of document reranking, which allows users to improve retrieval quality by reordering documents based on relevance before passing them to the model. On the core side, standard model exception types have been introduced, making it easier to handle model-related errors consistently across different providers.
Several bug fixes address edge cases that could affect Fireworks users. The integration now filters out invalid tool calls that may appear in version 1 content, preventing crashes when a model returns malformed tool invocations. It also avoids sending a repeated service_tier parameter, which could previously cause duplicate values in API requests.
Maintenance work included refreshing the internal model profile data twice to keep model metadata up to date. Dependency updates were applied: langsmith was bumped from 0.10.6 to 0.10.16, and aiohttp from 3.14.1 to 3.14.3, ensuring compatibility and incorporating upstream fixes.
