llama.cpp Releases
609 releases of ggml-org/llama.cpp
- b9768
This release introduces support for the Granite Speech Plus audio model, including conversion capabilities and extended multi-layer concatenation features. Various bug fixes were also applied to naming and alignment.
Jun 23, 2026
- b9767
This release optimizes WebGPU inference for small batches by utilizing a matrix-vector path. It includes updated binaries for supported platforms and backends.
Jun 23, 2026
- b9763
This release adds an identifier to tool call responses within the server API. Pre-built binaries are provided for macOS, Linux, Android, Windows, and openEuler across various architectures and hardware backends.
Jun 22, 2026
- b9761
Moved model downloading to a dedicated child process to allow for real-time progress tracking via SSE endpoint.
Jun 22, 2026
- b9760
Refactored and generalized the input file schema in the server to support video inputs and raw base64 encoding.
Jun 22, 2026
- b9758
Added BFloat16 (BF16) support for binary broadcast and unary operations in the SYCL backend.
Jun 22, 2026
- b9757
Removed the unconditional application of softmax and sorting in the top-n-sigma sampling algorithm.
Jun 22, 2026
- b9756
Fixed a heap-buffer-overflow crash in the server's edit file functionality when appending to the end of an empty file.
Jun 22, 2026
- b9754
Implemented an AC (Aho-Corasick) parser for the PEG grammar generation module to enforce stricter grammar rules.
Jun 21, 2026
- b9753
Fixed progress reporting for loading specification models and introduced a list of stages to track the loading process more accurately.
Jun 21, 2026
- b9752
Refactored the server's batch construction logic to handle batch overflow more carefully and added an abort mechanism for slots.
Jun 21, 2026
- b9751
This release fixes a bug in the Multi-Modal (mtmd) module where the memory usage calculation was incorrect. It provides the standard set of pre-built binaries for supported platforms.
Jun 21, 2026
- b9750
This release implements a new 'call statement' feature for Jinja templates, simplifying the implementation and reorganizing the caller context inside the function handler. It provides pre-built binaries for macOS, Linux, Android, Windows, and openEuler.
Jun 21, 2026
- b9748
This release adds a 'verbose' field to the server schema to improve logging verbosity control. It includes standard pre-built binaries for various operating systems and hardware backends.
Jun 21, 2026
- b9747
Adds real-time model loading progress tracking via a new Server-Sent Events (SSE) endpoint /models/sse in the server.
Jun 21, 2026
- b9745
Adds support for Flash Multi-Head Speculative Decoding (MTP3) for Step 3.5 and 3.7 models, including new APIs for layer offset management.
Jun 21, 2026
- b9744
Refactors GBNF (General Base-N Grammar) grammar generation into an Aho-Corasick automaton, improving performance and rule alignment.
Jun 21, 2026
- b9743
Aligns spacing rules in the JSON schema to grammar converter with the general parser formatting standards.
Jun 20, 2026
- b9742
Fixes a Hexagon DSP issue where padded strides were not used correctly for SSM (State Space Model) convolution weights.
Jun 20, 2026
- b9741
Refactors how quantization metadata is stored and accessed by using LLM_KV for quantization_version and file_type.
Jun 20, 2026