v1.6.2
langgenius/dify-sandboxv1.6.2Aug 1, 2026by amirivojdan
AI Summary
A maintenance release focused on enhancing security by shipping word embeddings in NumPy `.npz` format, improving error handling by raising `TypeError` for invalid inputs, and upgrading the development toolchain.
Key Highlights
- Safer Word Embeddings: Models ship in NumPy `.npz` format with pickling disabled
- Clearer Error Types: Unsupported inputs now raise `TypeError` instead of `ValueError`
- Code Quality: Upgraded development toolchain and cleaned up codebase issues
Breaking Changes
- Passing an unsupported input type to transforms or pipelines now raises `TypeError` instead of `ValueError`
New Features
- Word embedding models in NumPy `.npz` format
- Restricted unpickler for existing `.bin` models
- General code quality improvements
Full Release Notes
# What's New in v1.6.2 This is a minor maintenance release focused on safer word embedding loading and internal code quality. ## Highlights * **Safer Word Embeddings** – word embedding models now ship in NumPy's `.npz` format and load with pickling disabled, removing arbitrary code execution risk at load time. Existing `.bin` models continue to load through a restricted unpickler. * **Clearer Error Types** – passing an unsupported input type to a transform, pipeline, or word cloud now raises `TypeError` instead of `ValueError`. * **Code Quality** – upgraded the development toolchain and cleaned up the issues it surfaced across the codebase.