llama.cpp Releases
609 releases of ggml-org/llama.cpp
- b9967
Updated the LLM server API to accept null values for sampling parameters (like temperature and top_p), allowing clients to request server defaults and improving OpenAI API compliance.
Jul 12, 2026
- b9966
Optimized regex pattern compilation in `llama_meta_device_get_split_state()` by making patterns static const local variables to prevent recompilation on every call, significantly improving decode thread performance.
Jul 11, 2026
- b9965
Enhanced ARGSORT performance for small tensors on Hexagon hardware by implementing efficient bitonic sorting in HVX registers and optimizing common cases.
Jul 11, 2026
- b9964
Fixes an issue where duplicate downloads of spec models could occur in arguments.
Jul 11, 2026
- b9963
Adds multi-tile support for DeepSeek OCR v1, enabling dynamic resolution processing.
Jul 11, 2026
- b9960
Removes the `loading.html` file from the server interface.
Jul 11, 2026
- b9959
Syncs and updates the ggml library core.
Jul 11, 2026
- b9957
Improves server tools by adding abstractions and removes the deprecated `apply_diff` functionality.
Jul 10, 2026
- b9956
Fixes a crash in the CLI when connecting to a server with an invalid base URL by implementing graceful exception handling.
Jul 10, 2026
- b9952
Optimizes FlashAttention usage by using F16 for Key-Query masks and removes dead code related to raw_k repeats in DeepSeek V4.
Jul 10, 2026
- b9951
Introduces a new ET backend with FlashAttention and SSM support, along with optimized kernels for various operations like MUL_MAT, ROPE, and GLU.
Jul 10, 2026
- b9950
Introduces unit tests for the `llama-batch` system and fixes Windows build issues.
Jul 10, 2026
- b9949
Adds OpenCL cluster-parallel decode FlashAttention support for Adreno GPUs, optimizing performance on Android and Windows devices.
Jul 10, 2026
- b9948
Optimized CUDA top-k and argsort implementations to reduce temporary buffer memory usage.
Jul 10, 2026
- b9947
Added --output option to CLI for specifying output file path.
Jul 10, 2026
- b9946
Added optimizations for Hexagon (Qualcomm) backend including tiling and fast division.
Jul 9, 2026
- b9945
Fixed crash when using chat templates that fail at apply time but pass at init time.
Jul 9, 2026
- b9941
Optimized compilation by avoiding reliance on compile-time optimizations and using cheaper arithmetic operations.
Jul 9, 2026
- b9940
Added offline parameter support to the llama-bench benchmarking tool.
Jul 9, 2026
- b9939
Added depthwise convolution support for Metal backend and fixed F16 kernel support for the new operation.
Jul 9, 2026