v2.6.0 v2.6.1

medium risk

tulir/whatsmeowAug 27, 2026 → Aug 28, 20262 days between releases

What's New

Unified chat templating and retired legacy server in v2.6.1.

Breaking Changes

  • Retired legacy shimmy_server_gpu server

New Features

  • Unified chat templating using real GGUF templates
  • Offline routing gate for template validation
  • Extended shimmyjinja for slice-with-step support

Improvements

  • Updated Airframe dependency to version 0.4.1
  • Simplified chat prompt rendering logic

Bug Fixes

  • Fixed garbled output on Qwen3 and Gemma models

Upgrade Notes

Migrate away from the retired shimmy_server_gpu binary. Ensure all API endpoints use the new unified prompt rendering module.

The Two Releases

v2.6.0

Shimmy v2.6.0

Aug 27, 2026· 2 months ago

Consolidated the certification pipeline into a clean 3-box regimen and removed legacy engine code.

Full release notes
v2.6.1

v2.6.1

Aug 28, 2026· 2 months ago

Implemented a unified chat templating system using real GGUF templates and retired the legacy server.

Full release notes