b9874
ggml-org/llama.cppb9874Jul 5, 2026by github-actions[bot]1 days after b9873
What changed since b9873
b9874 adds a CUDA concat kernel for quantized tensors, following the K/V rotation crash guard in b9873.
New features
- CUDA backend now supports concat for quantized tensor types, mirroring the earlier CPU implementation.
Improvements
- Quantized tensor concatenation on GPU avoids CPU fallback, improving performance for CUDA workloads.
- Code for the new CUDA concat path was simplified per a contributor suggestion.
Upgrade notes
This is a routine incremental build with no config or API changes; existing flags and model files remain compatible. CUDA users doing quantized concat operations should see faster execution, while other backends are unaffected.
Analyzed Oct 30, 2026
Upgrading?
Release notes summary
Implements a CUDA version of the concat operation for quantized tensors. This mirrors the CPU fix from b9871 and improves GPU backend performance.
Key Highlights
- Adds CUDA concat implementation for quantized types
- Optimizes GPU workloads involving quantized tensor concatenation
- Applies code simplification suggestion from contributor
Release Notes (excerpt)
<details open> cuda : concat implementation for quantized types (#25303) * cuda : concat implementation for quantized types * chore : apply am17an clever suggestion to shorten the code --------- Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com> </details> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:** - Ubuntu x64 (CPU) - Ubuntu arm64 (CPU) - Ubuntu s390x (CPU)…
Published by ggml-org on GitHub.
Read the full release notesRelated Repositories
:electron: Build cross-platform desktop apps with JavaScript, HTML, and CSS
Godot Engine – Multi-platform 2D and 3D game engine
ClickHouse® is a real-time analytics database management system
Self-hosted game stream host for Moonlight.
DuckDB is a high-performance analytical database system