LlamaIndex v0.14.23
LlamaIndex v0.14.23 introduces multimodal synthesis improvements and query engines, adds DocumentBlock and VideoBlock parsing to FunctionTool, prevents state mutation leaks in Workflows, and fixes several core ingestion and text-splitting bugs.
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
LlamaIndex release v0.14.23 updates core and various integration packages. Key highlights include: - Core Features: Part 2 of multimodal synthesis, multimodal query engines, and a mock LLM for tool calling. - Tool & Agent Updates: FunctionTool parsing now supports DocumentBlock and VideoBlock; preserved URL-backed video and document memory blocks. - Workflow & State Fixes: Initial state is now deep-copied in workflows to prevent mutation leaks across consecutive runs. - Text Processing & Ingestion: Fixed RecursionError in TokenTextSplitter and SentenceSplitter when units exceed chunk size, prevented ZeroDivisionError in PromptHelper on empty inputs, enforced utf-8 encoding across text I/O, used running event loop in ingestion pipelines, and optimized batch deduplication using sets. - Metadata & Retrieval: Fixed NE and NIN metadata filters when matching missing fields, preserved TreeSelectLeafRetriever source nodes, and retained IndexNode objects during model dumps. - Ecosystem Updates: Supported google-genai 2.0+ SDKs in llama-index-embeddings-google-genai, alongside routine dependency bumps across multiple packages.
可执行摘要
Upgrade to llama-index-core 0.14.23 for multimodal synthesis updates, mock LLM tool calling, and critical fixes: deep copying Workflow initial state to avoid run leakage, handling larger-than-chunk-size units in TokenTextSplitter/SentenceSplitter to stop RecursionError, and supporting DocumentBlock/VideoBlock in FunctionTool parsing.
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade llama-index-core to 0.14.23.
- Verify workflows that rely on state to ensure existing runs were not inadvertently dependent on shared mutable state across executions.
- Update llama-index-embeddings-google-genai if adopting google-genai 2.0+ SDKs.