v0.0.21 @magnitudedev/cli@0.2.8
medium risk
facebook/astryxOct 12, 2026 → Oct 12, 20260 days between releases
What's New
This release adds streaming embeddings and GPU compatibility fixes, though the notes appear to be from a different repository than the older version.
New Features
- Streaming embeddings for large models to reduce memory usage.
- Gemma 4 models paired with assistant models for faster generation.
- GPU memory optimizations for AMD/NVIDIA/Apple Silicon.
Improvements
- Stop decoding zstd requests at the 128 MB routing limit to prevent memory expansion.
- Faster Gemma decoding after the first-load optimization.
- Improved long prompt handling on 8 GB GPUs.
Bug Fixes
- Fix Qwen 3.5, 3.6, and 3.8 models failing on M1 Max and M1 Ultra Macs.
- Fix AMD GPUs on Windows unloading the model after long prompts.
Upgrade Notes
No configuration changes are required. Ensure your GPU drivers are up to date for the AMD and Apple Silicon fixes.
The Two Releases
v0.0.21
Instatic 0.0.21
Oct 12, 2026· 10 days ago
A bug fix release addressing editor classes, settings modal behavior, and MCP OAuth discovery security.
Full release notes@magnitudedev/cli@0.2.8
@magnitudedev/cli@0.2.8
Oct 12, 2026· 9 days ago
This release focuses on memory management for large models and GPU compatibility fixes, including streaming embeddings and optimizations for Apple Silicon and AMD/NVIDIA cards.
Full release notes