llama.cpp Releases
609 releases of ggml-org/llama.cpp
- b9835
Fixes a UI bug where stop sequences and reasoning steps were incorrectly skipped in single-model mode.
Jun 28, 2026
- b9833
Implements a dedicated parser for MiniCPM5, refactoring tool-call handling and fixing Jinja API compatibility.
Jun 28, 2026
- b9832
Enhances the Jinja template engine with a debugging flag to output the generated program for troubleshooting.
Jun 28, 2026
- b9831
Introduces support for DFlash v2 architecture, including a new sliding window attention mechanism per layer type.
Jun 28, 2026
- b9830
Adds an offline mode to the `llama download` tool and fixes a critical use-after-free bug in the URL-task callback.
Jun 28, 2026
- b9829
This release focuses on reducing logging verbosity and performing code cleanup, specifically renaming constants and updating common components.
Jun 28, 2026
- b9828
Significant performance and feature updates to the OpenCL backend including a major rework of Flash Attention kernels supporting various quantizations and precisions.
Jun 27, 2026
- b9827
Optimized CUDA memory copy operations by adding a fast path for 2D strided tensor copies using cudaMemcpy2DAsync.
Jun 27, 2026
- b9826
Fixed unit tests related to normalization operations in the SYCL backend.
Jun 27, 2026
- b9825
Fixed a bug in the Vulkan backend where the step operator failed with zero inputs.
Jun 27, 2026
- b9824
Renamed RPC server binary to ggml-rpc-server and standardized naming for graph export tools.
Jun 27, 2026
- b9823
Added Windows OpenVINO builds to the CI check-release pipeline.
Jun 27, 2026
- b9822
Bug fix for the chat template test runner to correctly handle the --no-common option.
Jun 27, 2026
- b9821
Added command-line flags to display version, licenses, and help information.
Jun 26, 2026
- b9820
Performance improvements to the scheduler by reducing synchronization overhead during split compute and adding async CPU-to-GPU copy capabilities.
Jun 26, 2026
- b9817
Major update to the OpenVINO backend including version upgrade to 2026.2.1, removal of hardcoded compute types, and new operator support like GLU SwiGLU OAI.
Jun 26, 2026
- b9816
GGML synchronization updates to improve thread safety and consistency across the library.
Jun 26, 2026
- b9814
Optimizes the Vulkan backend's mul_mat_vecq operation for AMD Radeon MI50 GPUs, improving performance for quantized matrix multiplication operations.
Jun 26, 2026
- b9813
Adds support for Intel Xe1 architecture and enables cooperative matrix operations (coopmat1) on Intel Xe-LPG Plus GPUs, improving performance for Intel GPU users.
Jun 26, 2026
- b9811
Workaround for a compiler bug in the Vulkan backend's conv2d coopmat2 path, with the same fix applied to CONV_3D operations.
Jun 26, 2026