v0.5.8 @magnitudedev/cli@0.2.8

low risk

firecrawl/pdf-inspectorOct 12, 2026 → Oct 12, 20260 days between releases

What's New

Newer release focuses on memory management and GPU stability, while older release introduced a new file format and performance gains.

New Features

  • Streaming embedding tables to reduce memory usage.
  • Gemma 4 models paired with assistant models for faster generation.

Improvements

  • Stops decoding zstd requests after the 128 MB limit to prevent gigabyte memory expansion.
  • Performance optimizations for Apple Silicon and NVIDIA GPUs.
  • Streaming embedding tables reduce memory usage (e.g., Qwen3.8 27B saves ~600 MB).

Bug Fixes

  • Fixes Qwen 3.5, 3.6, and 3.8 crashing on M1 Max and M1 Ultra Macs.
  • Fixes long prompts failing on 8 GB AMD GPUs.

Upgrade Notes

Ensure models are compatible with the new zstd decoding limit (128 MB) to prevent memory spikes. Verify GPU compatibility for Qwen models on M1 Max/Ultra and AMD GPUs with 8GB VRAM.

The Two Releases

v0.5.8

IngeTrazo 0.5.8

Oct 12, 2026· 10 days ago

A major performance and feature update introducing an Extension Manager, document units, and a Model Browser, alongside significant speed improvements for large models and a new file format.

Full release notes
@magnitudedev/cli@0.2.8

@magnitudedev/cli@0.2.8

Oct 12, 2026· 9 days ago

A patch update focusing on memory management for zstd Codex requests and performance optimizations for Apple Silicon and NVIDIA GPUs, while fixing stability issues on Macs and AMD GPUs.

Full release notes