v0.76.0 v0.76.1

low risk

silvia-odwyer/photonSep 10, 2026 → Sep 12, 20262 days between releases

What's New

Patch release adds GPU diagnostics and fixes tool validation.

New Features

  • Itemized advertised capacity behind vram_bytes
  • Gpu_name_source display
  • Diagnose direct-connect advertisement and peer paths
  • Surface advertised memory breakdown

Improvements

  • Enhanced Linux CUDA support on driver-only hosts
  • Enhanced GPU memory reporting and breakdown
  • Improved diagnosis of direct-connect advertisements

Bug Fixes

  • Reject malformed tool definitions
  • Prevent self-identity joins
  • Keep direct GGUF split loads local
  • Match NVIDIA driver devices by identity

Upgrade Notes

v0.76.1 introduces detailed GPU memory reporting and diagnostics. Users should verify tool definitions are valid as the release now rejects malformed inputs. No configuration changes are required.

The Two Releases

v0.76.0

v0.76.0

Sep 10, 2026· 1 month ago

A major release introducing durable KV prefix cache, unified radix caching, and SafeTensors checkpoint support. It also enforces configuration schema parity and wires significant hardware placement and speculative decoding settings.

Full release notes
v0.76.1

v0.76.1

Sep 12, 2026· 1 month ago

This release focuses on bug fixes and diagnostics improvements, including enhanced GPU memory reporting, direct-connect path diagnosis, and fixes for malformed tool definitions and self-identity joins.

Full release notes