llama.cpp Releases
609 releases of ggml-org/llama.cpp
- b10218
Adds downsampling support for MiniCPMv46 multimodal models, including integration into the image preprocessor.
Aug 1, 2026
- b10217
Enables tool calling capabilities within the 'thinking' process for DeepSeek-V3/4 (DS4) models in the chat interface.
Aug 1, 2026
- b10216
Adds support for the POOL_1D operation to the Vulkan backend, including a compute shader and fixes for boundary crash issues.
Jul 31, 2026
- b10215
Fixes a crash on Windows Intel GPUs by introducing a driver version check to ensure compatibility with driver version 32.0.101.8860 and above.
Jul 31, 2026
- b10214
Adds n_embd_head support to multimodal models and includes builds for various platforms including macOS, Linux, Windows, and Android.
Jul 31, 2026
- b10213
This release adds support for quantizing rotated Key-Value (KV) caches, which is relevant for models utilizing specific cache layouts.
Jul 31, 2026
- b10212
This release optimizes model loading by skipping the loading of MTP (Multi-Token Prediction) tensors if they are not used by the current model configuration.
Jul 31, 2026
- b10211
This release updates the Vulkan SDK to version 1.4.357.0 to ensure compatibility with the latest hardware and driver features.
Jul 31, 2026
- b10210
This release fixes a bug in the server API where accepted tokens were not correctly counted when draft token replay was needed during generation.
Jul 31, 2026
- b10209
This release optimizes CUDA kernel performance for Q2_0 quantization by using the __byte_perm intrinsic for efficient element extraction.
Jul 31, 2026
- b10208
This major release introduces oneMKL GEMM-based Flash Attention for SYCL, delivering up to 2x performance improvements on XMX-accelerated hardware (Intel Arc/Battlemage). It includes fixes for multi-turn conversation corruption and stride handling.
Jul 31, 2026
- b10207
This release fixes SYCL support for data copy operations by adding support for previously missed data types and cleaning up unused code.
Jul 31, 2026
- b10206
This release enforces consistent Key and Value cache types for DeepSeek V4 and MLA models, and enables Flash Attention automatically when the Value cache is quantized.
Jul 31, 2026
- b10205
This release focuses on optimizing group matrix multiplication for the Zendnn backend, specifically targeting the mul_mat_id operation for improved efficiency.
Jul 31, 2026
- b10204
This release adds support for device-to-device memory copying in SYCL using the DEV2DEV_MEMCPY_FORWARD mechanism, optimizing performance for Intel hardware.
Jul 31, 2026
- b10203
This release adds support for the Q2 quantization format in matrix multiplication operations within the SYCL backend.
Jul 31, 2026
- b10202
The SYCL backend has been optimized by fusing the RMS Normalization and Multiply operations into a single kernel.
Jul 31, 2026
- b10201
This release optimizes the WebGPU implementation of Flash Attention for quantized Key-Value (KV) caches, specifically improving performance at long context lengths.
Jul 31, 2026
- b10199
The HTTP server has been updated to support embedding inputs for generating the next token, improving the server's input handling capabilities.
Jul 30, 2026
- b10198
This release expands Vulkan backend capabilities by adding support for quantized concatenation operations.
Jul 30, 2026