●Change8

v1.79.1.dev5

📦 litellmView on GitHub →
✨ 11 features🐛 12 fixes🔧 12 symbols

Summary

This release introduces significant feature enhancements, including Milvus vector store support and expanded Passthrough API capabilities for Azure AI Vector Stores. It also addresses several critical bugs related to memory usage, AWS Bedrock, and model configuration.

Migration Steps

  1. If you are using Pydantic v2.11+, ensure your environment is configured to handle deprecation warnings to prevent potential memory accumulation.
  2. Review minimum resource requirements for production environments as they have been added.

✨ New Features

  • Milvus vector store search support added.
  • Passthrough Endpoints now support multi-part form data.
  • Azure AI Vector Stores support for "virtual" indexes and vector store creation via passthrough API.
  • Added custom_llm_provider support for non-generation video endpoints.
  • LLM provider response headers are now included in Responses API.
  • E2E Container API Support added.
  • Guardrail information is now expected to be a list type.
  • Milvus Passthrough API support added for creating and reading vector stores.
  • UI: Added functionality to delete Team Members with friction.
  • UI: Improved existing credentials handling for Models.
  • UI: Guardrail Info Page now displays PII Configuration.

🐛 Bug Fixes

  • Resolved memory accumulation issue caused by Pydantic 2.11+ deprecation warnings.
  • Fixed empty assistant message handling in AWS Bedrock Converse API to prevent 400 Bad Request errors.
  • Fixed issue where Azure did not accept extra body parameters.
  • Fixed Anthropic token counting for VertexAI.
  • Base64 strings are now stripped in S3 operations.
  • Fixed regression in UI where Guardrail Entity could not be selected and was not displayed.
  • Fixed support for dot notation in ui_sso.py.
  • Langfuse OTEL now supports streaming responses API.
  • The diarized transcript is now returned when required in the request.
  • Fixed index field not being populated in streaming mode when n>1 and tool calls are present.
  • Updated perplexity cost tracking.
  • Fixed image_config.aspect_ratio not working for gemini-2.5-flash-image.

🔧 Affected Symbols

Passthrough EndpointsAzure AI Vector StoresAWS Bedrock Converse APIResponses APIAnthropic token countingVertexAIs3ui_sso.pylangfuse otelguardrail_informationimage_config.aspect_ratiogemini-2.5-flash-image