Instructor v1.15.3
Instructor v1.15.3 brings multiple fixes including sensitive key redaction in debug logs, unwrapped propagation of IncompleteOutputException, and improvements for Anthropic, GenAI/VertexAI, and responses tools modes.
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
Instructor v1.15.3 introduces various bug fixes, typing improvements, and dependency updates. Key changes include: redacting API keys and sensitive parameters in response model debug logs; letting IncompleteOutputException propagate directly without wrapping on retry; excluding None fields when serializing Anthropic tool-use blocks during reask; skipping non-text parts in GenAI/VertexAI templating; preserving descriptions and surfacing reasoning summary events in responses tools mode; increasing the litellm upper bound to 1.83.7; and multiple refactors and typing coverage improvements related to v2 migration.
可执行摘要
Instructor release v1.15.3 addresses sensitive data logging, unwraps IncompleteOutputException during retries, resolves Anthropic tool-use serialization issues on reasks, and updates litellm constraints along with v2 refactors.
- 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 instructor to v1.15.3 via your package manager.
- Verify exception handling around retries if relying on previous wrapped behavior for IncompleteOutputException.
- Verify tool re-asking when using Anthropic models to ensure serialization works as expected.