v0.2.1 @magnitudedev/cli@0.2.5
bespokelabsai/nimbleOct 3, 2026 → Oct 3, 20260 days between releases
What's New
Performance optimizations for MoE models and fixes for Metal kernel compilation on M1/M2 Macs
New Features
- Multi-token prediction speedups on NVIDIA GPUs
- 2x faster prompt processing on M5 and later Macs
- Metal kernel compatibility fixes for M1 and M2 Macs
- Vulkan GPU shader compilation fixes
Improvements
- Optimized MoE expert weight loading for multi-token prediction
- Context window speedups (16K) on Macs and NVIDIA GPUs
- GPU tensor operations for prompt processing on M5+ Macs
Bug Fixes
- Metal kernel compilation errors on M1 and M2 Macs
- Models with 16 or more query heads failing on M1/M2
- Vulkan GPU shader compilation issues
Upgrade Notes
The provided release notes are for different repositories (nimble vs magnitude-cli), so there are no upgrade notes between them.
The Two Releases
v0.2.1 — God's Eye View through Your Agent (with security fixes and other features)
Oct 3, 2026· 19 days ago
Focuses on security hardening, including throttling cost-bearing proxies and improving Host checks, alongside a smaller build size and Cyber HUD layout fixes.
Full release notes@magnitudedev/cli@0.2.5
Oct 3, 2026· 19 days ago
Focuses on performance optimizations for Mixture-of-Experts models and fixes compatibility issues with Metal kernels on M1/M2 Macs and Vulkan GPUs.
Full release notes