llama.cpp Releases
609 releases of ggml-org/llama.cpp
- b9810
Adds cublasSgemmBatched mapping for HIP/MUSA vendor headers, improving cross-platform compatibility between CUDA and HIP/MUSA.
Jun 26, 2026
- b9803
Improves OpenCL profiling by flushing incomplete batches at shutdown, ensuring profiling data is captured even when processing is interrupted.
Jun 26, 2026
- b9802
Minor release with no specific changes documented in the release notes.
Jun 25, 2026
- b9789
Fixes a quantization bug affecting Mixture of Experts (MoE) models when using MTP (Mixture of Token Pruning).
Jun 25, 2026
- b9788
Adds SYCL tensor parallelism support for dual-GPU setups, implementing a BF16-compressed all-reduce path for large tensors and achieving significant performance improvements on Intel Arc Pro B70.
Jun 25, 2026
- b9787
Fixes unit test failures for 3D convolution operations in the SYCL backend.
Jun 25, 2026
- b9786
Enhances the OpenCL backend to support non-contiguous memory rows in normalization operations, improving compatibility with various memory layouts.
Jun 25, 2026
- b9785
Hardens the chat mode input handling to improve security by enforcing stricter checks on input caps.
Jun 25, 2026
- b9784
Major performance and architectural overhaul of the Hexagon (Qualcomm) backend, introducing tiled weight repacking (32x32), improved kernel parameters, graph caching, and VTCM optimizations.
Jun 24, 2026
- b9782
Code cleanup removing the unused `json-partial` module from the common library.
Jun 24, 2026
- b9781
Adds a configuration option to reduce the graph submission batch size in the Vulkan backend to prevent timeout issues on certain hardware configurations.
Jun 24, 2026
- b9780
Improves build reliability for the Vulkan backend by ensuring the build fails immediately if a shader fails to compile, rather than silently succeeding and producing broken binaries at runtime.
Jun 24, 2026
- b9777
Adds two new LFM2.5 embedding models to the repository, specifically LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M, and restores LFM models in the README.
Jun 24, 2026
- b9776
Fixes a Vulkan floating-point overflow issue in Fully Attention (FA) by applying bias before the softmax operation. Includes standard binary builds for macOS, Linux, Windows, Android, and openEuler.
Jun 24, 2026
- b9775
The server component now includes a check for draft context creation errors to improve error reporting and debugging.
Jun 23, 2026
- b9774
This release enhances Vulkan backend support by implementing unary operations and non-contiguous memory support for various math functions like SQR, SQRT, and NORM.
Jun 23, 2026
- b9773
This release adds support for the GET_ROWS_BACK operation in the Vulkan backend.
Jun 23, 2026
- b9771
This release optimizes Vulkan shader variants by converting the ALIGNED flag for matrix multiplication into a specialization constant, which reduces binary size.
Jun 23, 2026
- b9770
The server component receives a fix for remote preset handling to ensure correct operation. A test was also added to validate this functionality.
Jun 23, 2026
- b9769
This release fixes a linker error in Vulkan builds by explicitly linking the ggml-cpu library when debug result-checking or test modes are enabled. Standard binaries are provided for all supported platforms.
Jun 23, 2026