Release notes from llama.cpp

https://github.com/ggml-org/llama.cpp/releases.atom · nguồn: rss


b10872

2026-09-09T07:47:26.000Z · github-actions[bot]

jinja: treat a null left operand of in as a plain lookup (#28620) Templates that default an optional variable to none and then test its Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/46177710 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu x64 (CPU) Ubuntu arm6

b10871

2026-09-09T07:20:12.000Z · github-actions[bot]

vulkan: add dedicated iq4_xs mat-vec shader (#28426) vulkan: add dedicated iq4_xs mat-vec shader Dedicated mul_mat_vec_iq4_xs for the dmmv path, replacing the generic fallback. ~+6-17% token generation on RDNA4 depending on model. Assisted-by: Pi agent with Qwen3.8 27B vulkan iq4_xs: remove dead n_it unroll branch Remove the n_it <= 8 experimental branch that attempted to fully unroll Website: htt

b10870

2026-09-09T06:40:30.000Z · github-actions[bot]

vulkan: add f16 B-type matmul pipelines and warp tile size tuning for Intel coopmat1 (#27471) vulkan: add f16 B-type matmul pipelines and warp tile size tuning for Intel coopmat1 simplify mmp selection in mul_mat_id per review comment vulkan: enable f16 B-type pipeline for dense matmul on all vendors (MoE stays Intel-only) vulkan: add missing ocp_fp4 branches, restrict required_subgroup_size to In

b10869

2026-09-09T05:27:37.000Z · github-actions[bot]

tests : use less threads for data initialization (#28325) tests : use 1 thread for data initialization cont : scale threads with number of elements cont : adjust Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/46152521 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ub

b10868

2026-09-09T02:03:08.000Z · github-actions[bot]

Add IQ type handling for MoE (#28476) Co-authored-by: cwriter cwriter@localhost Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/46124981 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu x64 (CPU) Ubuntu arm64 (CPU) Ubuntu s390x (CPU) Ubuntu x64 (Vulkan) Ubuntu ar

b10867

2026-09-08T17:33:23.000Z · github-actions[bot]

llama: disable lazy tensor loading by default on iGPUs (#28326) llama: add lazy mode auto, fix iGPU regression revert changes except disabling lazy load on iGPUs in AUTO Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/46033570 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework L

b10865

2026-09-08T15:14:21.000Z · github-actions[bot]

Revert "ggml-cuda : restore prop.integrated on HIP builds (#24233)" (#28604) This reverts commit c7d8722. Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/46002046 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu x64 (CPU) Ubuntu arm64 (CPU) Ubuntu s390x (CPU) Ubu

b10864

2026-09-08T14:33:11.000Z · github-actions[bot]

server : apply checkpoint min-step eviction only when the checkpoint list is full (#28302) The spacing eviction in create_checkpoint() keeps the oldest checkpoint and Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/45991527 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linu

b10863

2026-09-08T14:01:19.000Z · github-actions[bot]

metal : fix idle threads in mul_mv_iq3_xxs for ne00 < 1024 (#28086) metal : fix half-idle simdgroup in kernel_mul_mv_iq3_xxs_f32 for ne00 < 1024 metal : keep N_R0_IQ3_XXS = 4, dispatch a separate 8-row split kernel for ne00/32 < 32 The plain kernel is unchanged from master (4 rows per simdgroup, one thread per metal : select the iq3_xxs row split with a function constant instead of a separate kern

b10859

2026-09-08T12:36:49.000Z · github-actions[bot]

llama : add missing headers (#28566) fix compile-error: add missing header Bug: #28557 [email protected] fix compile-error: add missing header Bug: #28559 [email protected] fix compile-error: add missing header Bug: #28560 [email protected] fix compile-error: add missing header Bug: #28561 [email protected] fix compile-error: add missing header Bug: #28562 [email protected] fix comp