LiteLLM v1.103.0-rc.1
LiteLLM v1.103.0-rc.1 introduces 429 rejections for saturated deployment slots, Microsoft Foundry agent connectivity via Entra auth, Amazon Transcribe passthrough, and various proxy performance and routing fixes.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
LiteLLM v1.103.0-rc.1 is a release candidate containing numerous proxy, routing, and provider integration updates. Key additions include 429 status rejections when deployment max parallel request slots are exhausted, integration with Microsoft Foundry agents via Entra auth, support for Amazon Transcribe pass-through with completion-time job pricing, live MCP gateway session visibility, and separate login/secret namespaces for HashiCorp Vault. The release also shifts prompt injection heuristics off the main event loop, ensures config files take precedence over database configurations, evicts JWT key mapping caches on entity deletion, optimizes startup times by deferring FastAPI and Tiktoken imports, and resolves various pricing and cost calculation edge cases across Mistral, Bedrock Titan, and OpenRouter.
可执行摘要
LiteLLM release v1.103.0-rc.1 adds router-level 429 rejections when max parallel request slots fill, integration for Microsoft Foundry agents via Entra auth, live MCP gateway session tracking, off-event-loop prompt injection heuristics, and deferred imports of fastapi and tiktoken. Configuration resolution was refactored so config files supersede the database.
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Verify the container image using Cosign against key commit 0112e53 or tag v1.103.0-rc.1 if deploying via Docker.
- Review deployment routing configurations to account for immediate 429 rejections when max parallel request slots are saturated.
- Ensure deployment settings reflect the new precedence rule where config file definitions override database values.