ggml-org/llama.cpp

Pull Requests Commits

0cc4m committed 6 days ago

bad0e3e9

tune implementation

0cc4m committed 6 days ago

6cf0766d

vulkan: add sparse Flash Attention support for DSV4/GLM

0cc4m committed 6 days ago

48c5d442

remove sparse hint

am17an committed 7 days ago

a9bed298

am17an committed 8 days ago

bc03bdcc

CUDA + ggml: add sparse flash attention

am17an committed 8 days ago

17417268

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

kurquhar committed 9 days ago

Verified d7bd3bfc

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

ngxson committed 9 days ago

Verified 50f068ff

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

ngxson committed 9 days ago

Verified 6fe74980

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

Eric-A-Stalee committed 9 days ago

Verified b387ddfd

ggml : fix conv_transpose_2d for multiple batches (#26132)

tekinertekin committed 9 days ago

Verified a43c3986

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

ravel7524 committed 9 days ago

Verified 90c26fcd

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

ggerganov committed 9 days ago

Verified 86632248

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

Strongtut committed 9 days ago

Verified f5e85d43

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

wine99 committed 9 days ago

Verified 511f9c13

convert: prevent ndarray conversion in LazyChunkedTensor (#27869)

ngxson committed 9 days ago

Verified b19cbe92

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

johnkarlhill committed 9 days ago

Verified d077b4c2

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

Titaniumtown committed 9 days ago

Verified be876204

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

ggerganov committed 9 days ago

Verified 8963a9bd

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

infinitewarp committed 9 days ago

Verified 6d6b697c

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

ggerganov committed 9 days ago

Verified 4e97ac86

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

ruixiang63 committed 10 days ago

Verified ca3d5a3e

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

cqderek committed 10 days ago

Verified e70802a0

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

aparmp-quic committed 10 days ago

Verified 83d855c5

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

bartowski1182 committed 10 days ago

Verified 18443257

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

harkgill-amd committed 10 days ago

Verified 32176338

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

danielhanchen committed 10 days ago

Verified 6c84c7d5

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

slojosic-amd committed 10 days ago

Verified 6fdd0ac8

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

ngxson committed 10 days ago

Verified b10f9ca5

opencl: add bin kernels `kernel_gemm_moe_q4_0_q8_1_dp4a_bin`, `kernel_gemm_moe_mxfp4_q8_1_dp4a_bin` (#27768)

shawngu-quic committed 10 days ago

Verified 58546250

Older