llama.cpp Releases
684 releases of ggml-org/llama.cpp
- b9713
Enhances the Multimodal CLI by adding batching capabilities and introduces new video tests for better coverage.
Jun 19, 2026
- b9712
A maintenance release that fixes a build issue preventing the UI from compiling when the source directory is set to read-only.
Jun 19, 2026
- b9711
Refactors the handling of overview images in the Multimodal LLaVA-UHD model to standardize the processing pipeline.
Jun 19, 2026
- b9707
This release introduces schema support and validation for the server API, enhancing data integrity and error reporting capabilities.
Jun 19, 2026
- b9704
A build release fixing server error handling for invalid grammar.
Jun 18, 2026
- v1.15.0
Adds Watch Mode, GitHub shorthand auto-detection, and a token budget guard to Repomix.
Jun 18, 2026
- b9703
A build release reworking the router's handling of Hugging Face preset repositories.
Jun 18, 2026
- b9702
A build release fixing router arguments not being forwarded to child instances.
Jun 18, 2026
- b9701
A build release refactoring the multimodal preprocessor and adding new output structures.
Jun 18, 2026
- b9700
A build release renaming SYCL support macros for Level Zero API.
Jun 18, 2026
- b9699
A build release adding SYCL support for Q1_0 quantization operations.
Jun 18, 2026
- b9698
A build release modifying self-update behavior to only work when built with llama-install.sh.
Jun 18, 2026
- asr-models-qnn-2
Jun 18, 2026
- asr-models-qnn-binary-2
Jun 18, 2026
- b9697
A build release fixing the check-release message parsing logic.
Jun 18, 2026
- b9694
A build release fixing the Windows x64 OpenVINO release link.
Jun 18, 2026
- b9693
Adds a specific check for BF16 support within the Metal concat kernel to ensure proper functionality.
Jun 18, 2026
- b9692
Fixes the LLaVA-UHD multimodal model to no longer use the batch dimension, correcting a dimensional inconsistency.
Jun 17, 2026
- b9691
Conditionally enables the POWER11 backend based on compiler support flags to prevent build failures on current toolchains.
Jun 17, 2026
- b9690
Implements the backward RoPE (Rotary Positional Embedding) operator for the Metal backend by reusing existing kernels with a toggle function.
Jun 17, 2026