llama.cpp Releases
684 releases of ggml-org/llama.cpp
- b10428
Personal home directory paths have been replaced with generic placeholders in documentation and test fixtures to prevent information leakage.
Aug 14, 2026
- b10427
The SYCL backend performance is further enhanced by fusing MulMat (gate) + MulMat (up) + GLU operations for Q4_K dense FFN.
Aug 14, 2026
- b10426
The ggml library now forces single-thread execution on the WASI platform to ensure compatibility.
Aug 14, 2026
- b10425
SYCL backend performance is improved by fusing the gated-delta-net state writeback Copy operation.
Aug 14, 2026
- b10424
The DFlash module output logging for `target_layer_ids` has been clarified to improve readability for users.
Aug 14, 2026
- b10423
This release ensures that CPU parameters are consistently applied across all tools, maintaining configuration uniformity.
Aug 13, 2026
- b10419
A major update to the OpenVINO backend featuring Qwen3.5 support, significant memory optimizations to reduce compile-time peaks, and expanded operator support.
Aug 13, 2026
- b10418
The SYCL backend gains support for host pinned memory to improve Host-to-Device memory access, accompanied by a thread-safety fix.
Aug 13, 2026
- b10417
This release fixes an ambiguity in the LFM2 tool call argument name prefix within the chat module. It also includes standard builds for various operating systems and hardware backends.
Aug 13, 2026
- b10416
Fixes the web server to serve index.html with no-cache headers, ensuring users always receive the latest UI build.
Aug 13, 2026
- b10415
Adds automatic detection of MTP (Multi-Token Prediction) draft model types.
Aug 13, 2026
- b10414
Adds support for the TQ2_0 (ternary 2-bit) quantization type to the Metal backend and optimizes the mul_mv kernel.
Aug 13, 2026
- b10413
Adds automatic detection of speculative decoding types (dspark and dflash) when loading local draft models from GGUF metadata.
Aug 13, 2026
- b10412
Enables backend sampling support for speculative decoding, allowing it to work with both dflash and dspark models.
Aug 13, 2026
- b10411
Optimizes the Flash-Attention implementation by vectorizing the V-cache F16 to F32 conversion process.
Aug 13, 2026
- b10410
Refactors the SYCL gemm (General Matrix Multiply) implementation to use automatic fp16 promotion instead of manual type promotion.
Aug 13, 2026
- b10408
Adds support for SYCL ESIMD (Exponential-SIMD) kernels for multiple quantization types, improving performance on Intel GPUs.
Aug 13, 2026
- b10405
Removes unsafe math optimizations from HIP builds to ensure IEEE-conformant behavior and prevent greedy argmax divergence on RDNA3.5.
Aug 13, 2026
- b10400
Fixes for ARM builds and compiler warnings. Includes a fix for unused variables in the ggml library.
Aug 13, 2026
- b10375
Refined the chat parser for Qwen models to be more strict regarding bare function parsing to improve compatibility and robustness.
Aug 12, 2026