v2.2.3 @magnitudedev/cli@0.2.4
bespokelabsai/nimbleOct 1, 2026 → Oct 2, 20260 days between releases
What's New
Newer release optimizes tuning time, reduces memory, and fixes Mac SSH issues vs older naming consistency update
New Features
- Explicit Retry-After headers for failed requests
- Reduced kernel tuning time for large models
- Lower memory footprint for Qwen3.5-4B model
Improvements
- Kernel tuning time capped at 1 minute for large models
- Memory footprint reduced for Qwen3.5-4B model
- Retry-After headers added for failed requests
Bug Fixes
- SSH startup issues on Macs fixed
- Requests returning empty replies on partial failures fixed
- Models failing to load on M1/M2 Macs fixed
Upgrade Notes
No configuration changes required. Behavior changes include capping tuning time and returning 503 errors instead of empty replies for partial failures.
The Two Releases
v2.2.3
Oct 1, 2026· 20 days ago
This is a naming and release-reliability update ensuring the project is consistently referred to as 'ECC' everywhere, starting with the `pi/core` profile, and fixes the release workflow to wait longer for npm to serve artifacts.
Full release notes@magnitudedev/cli@0.2.4
Oct 2, 2026· 20 days ago
Optimizes kernel tuning time to a fixed minute and reduces memory overhead for models, alongside fixes for SSH startup issues and silent failure reporting.
Full release notes