v1.811.0 v0.76.1
NVIDIA/NeMo-Speech.cppSep 12, 2026 → Sep 12, 20260 days between releases
What's New
Newer release focuses on GPU diagnostics and cross-platform support, while the older release addressed Snowflake OAuth and Git sync issues.
Breaking Changes
- Match NVIDIA driver devices by identity instead of index
New Features
- Itemize the advertised capacity behind vram_bytes
- Add gpu_name_source to display how the value was captured
- Diagnose direct-connect advertisement and peer paths
- Surface the advertised memory breakdown
- Report client mode from --client --auto console state
Improvements
- Added support for Linux CUDA on driver-only hosts
- Fixed materialized cache writes on Windows to ensure stability
- Rejected malformed tool definitions and self-identity joins to enhance security
- Link Apple OpenMP runtime
- Keep direct GGUF split loads local
Bug Fixes
- Reject malformed tool definitions
- Link Apple OpenMP runtime
- Advertise externally observed public address
- Reject self-identity joins and add node key override
- Keep direct GGUF split loads local
Upgrade Notes
These releases originate from different repositories and are not compatible upgrades. The newer version changes GPU device matching from index-based to identity-based, which may affect configurations relying on specific device indices.
The Two Releases
v1.811.0
Sep 12, 2026· 1 month ago
This release improves database integration by enabling Snowflake OAuth as a dbt warehouse and fixes several issues related to Git sync, script deployment, and resource ownership.
Full release notesv0.76.1
Sep 12, 2026· 1 month ago
This release focuses on enhancing diagnostics for GPU memory and connectivity, improving cross-platform support for Linux CUDA and Windows, and fixing various stability and security issues within the inference framework.
Full release notes