ggml-org/llama.cpp

Pull Requests Commits

opencl: use a better matmul path on two Adreno GPU generations (#27640)

wanghqc committed 3 hours ago

Verified c841aeeb

metal : assert shared memory padding (#27951)

ggerganov committed 6 hours ago

Verified 3173a564

metal : add remaining fa-vec tunings for M4 Pro (#27915)

nikwen committed 8 hours ago

Verified 17252c76

sycl: make --fit respect --fit-target better (#27629)

nicois committed 12 hours ago

Verified cc83d7b4

vulkan: combine duplicated fastdiv functions, rename the one optimizing small divs (#27526)

jeffbolznv committed 13 hours ago

Verified c9ca51c1

metal : add fa-vec tunings for M1 Max (#27932)

jhen0409 committed 14 hours ago

Verified 5ea1b124

vulkan: Change mul_mat_id to pad K rather than N (#27925)

jeffbolznv committed 14 hours ago

Verified 77f132cb

snapdragon: python SDK setup (Windows) (#27903)

kurquhar committed 1 day ago

Verified d7bd3bfc

bench: add --tensor-read-lazy (#27881)

ngxson committed 1 day ago

Verified 50f068ff

model: qwen4exp: reduce number of graph splits (#27880)

ngxson committed 1 day ago

Verified 6fe74980

vulkan: fix missing view-alias dependencies in ggml_vk_graph_optimize (#27812)

Eric-A-Stalee committed 1 day ago

Verified b387ddfd

ggml : fix conv_transpose_2d for multiple batches (#26132)

tekinertekin committed 1 day ago

Verified a43c3986

Vulkan: add hoisting support for row IDs and expert count in shaders (#26686)

ravel7524 committed 1 day ago

Verified 90c26fcd

context : disable non-fused GDN and LID ops (#27877)

ggerganov committed 1 day ago

Verified 86632248

metal : add fa-vec tunings for M4 (#27875)

Strongtut committed 1 day ago

Verified f5e85d43

OpenVINO: Update OV to 2026.3.1, whisper.cpp support, Qwen3.5 on NPU, and new ops (#27843)

wine99 committed 1 day ago

Verified 511f9c13

convert: prevent ndarray conversion in LazyChunkedTensor (#27869)

ngxson committed 1 day ago

Verified b19cbe92

sycl: use TILE for quantized KV decode on BMG (#26689)

johnkarlhill committed 1 day ago

Verified d077b4c2

sycl: bind the f16 KV cache in place for the oneDNN SDPA path (#27468)

Titaniumtown committed 1 day ago

Verified be876204

metal : add fa-vec tunings for M3 Max, M5 and M5 Pro (#27863)

ggerganov committed 1 day ago

Verified 8963a9bd

metal : add fa-vec tunings for M4 Pro (#27824)

infinitewarp committed 1 day ago

Verified 6d6b697c

tests : run test-save-load-state across all architectures (#27755)

ggerganov committed 1 day ago

Verified 4e97ac86

model: add DSpark support for Nemotron3.5 (#27804)

ruixiang63 committed 1 day ago

Verified ca3d5a3e

ggml-hexagon: add HTP unary ops for ABS and LOG (#27786)

cqderek committed 1 day ago

Verified e70802a0

hex-unary: fix RMS_NORM_MUL weight-offset bugs for grouped/broadcast norms (#27798)

aparmp-quic committed 1 day ago

Verified 83d855c5

server: add ctx-per-slot (--kv-unified-per-slot) (#24124)

bartowski1182 committed 2 days ago

Verified 18443257

ci : build only the ggml-hip backend for windows-rocm release (#27753)

harkgill-amd committed 2 days ago

Verified 32176338

model: add Qwen3.8-Flash-Next (qwen4exp) (#27742)

danielhanchen committed 2 days ago

Verified 6c84c7d5

ci : bundle HIP runtime DLLs with Windows ROCm release (#26973)

slojosic-amd committed 2 days ago

Verified 6fdd0ac8

spec : add DFlash2 support (local convolution + candidate selector) (#27342) (#27816)

ngxson committed 2 days ago

Verified b10f9ca5

Older