ggml-cpu: disable GGML_NNPA by default due to instability#14879

Closed

taronaeo wants to merge 45 commits intoggml-org:masterfrom

taronaeo:fix/default-off-nnpa-simd

Commits on Jul 21, 2025

Commits on Jul 25, 2025

vulkan/cuda: Fix im2col when KW!=KH (#14789)

jeffbolznv
authored and
taronaeo
committed
docs : fix backends table in README.md (#14796)

rgerganov
authored and
taronaeo
committed
kleidiai: add support for get_rows (#14676)

chaxu01
authored and
taronaeo
committed
sycl: Fix im2col (#14797)

Rbiessy
authored and
taronaeo
committed
opencl: add conv2d kernel (#14403)

rmatif
authored and
taronaeo
committed
opencl: fix `im2col` when `KW!=KH` (#14803)

CISC
authored and
taronaeo
committed
cuda: remove linking to cublasLt (#14790)

yeahdongcn
authored and
taronaeo
committed
server : allow setting `--reverse-prompt` arg (#14799)

MollySophia
authored and
taronaeo
committed
opencl: remove unreachable `return` (#14806)

lhez
authored and
taronaeo
committed
cuda : implement bf16 cpy ops and enable bf16 cont (#14763)

CISC
authored and
taronaeo
committed
Mtmd: add a way to select device for vision encoder (#14236)

stduhpf
authored and
taronaeo
committed
imatrix: add option to display importance score statistics for a given imatrix file (#12718)

authored and
taronaeo
committed
llama : add model type detection for rwkv7 7B&14B (#14816)

MollySophia
authored and
taronaeo
committed
vulkan: fix rms_norm_mul to handle broadcasting dim0 (#14817)

jeffbolznv
authored and
taronaeo
committed
ggml : model card yaml tab->2xspace (#14819)

csabakecskemeti
authored and
taronaeo
committed
CUDA: add fused rms norm (#14800)

am17an
authored and
taronaeo
committed
CANN: weight format to NZ for Ascend310P3 (#14407)

tqgy6
authored and
taronaeo
committed
ggml: fix loongarch quantize_row_q8_1 error (#14827)

lixing-star
authored and
taronaeo
committed
memory : handle saving/loading null layers in recurrent memory (#14675)

authored and
taronaeo
committed
tests : add non-cont K,V FA tests

ggerganov
authored and
taronaeo
committed
CUDA: fix quantized KV cache + multiple sequences (#14822)

authored and
taronaeo
committed
ci : correct label refactor->refactoring (#14832)

CISC
authored and
taronaeo
committed
CUDA: fix compilation with GGML_CUDA_F16 (#14837)

JohannesGaessler
authored and
taronaeo
committed
CUDA: fix overflow in FA, tune performance (#14840)

JohannesGaessler
authored and
taronaeo
committed
convert : text-only support for GLM-4.1V-9B-Thinking (#14823)

jacekpoplawski
authored and
taronaeo
committed
sycl: fix undefined variable in work group size check (#14843)

djeong20
authored and
taronaeo
committed
metal : fix fusion across different encoders (#14849)

ggerganov
authored and
taronaeo
committed
docs: add libcurl-dev install hint for Linux distros (#14801)

authored and
taronaeo
committed
llama : fix MiniCPM inference after Granite Four changes (#14850)

jk3456a
authored and
taronaeo
committed
sycl: fixed semantics of block offset calculation (#14814)

Alcpz
authored and
taronaeo
committed
chat : fix kimi-k2 chat template (#14852)

ngxson
authored and
taronaeo
committed
context : perform output reorder lazily upon access after sync (#14853)

ggerganov
authored and
taronaeo
committed
ggml-cpu : remove stdlib include from repack.cpp (ggml/1276)

danbev
authored and
taronaeo
committed
cmake : fix usage issues (ggml/1257)

dg0yt
authored and
taronaeo
committed
sync : ggml

ggerganov
authored and
taronaeo
committed
musa: upgrade musa sdk to rc4.2.0 (#14498)

yeahdongcn
authored and
taronaeo
committed
sched : fix multiple evaluations of the same graph with pipeline parallelism (#14855)

slaren
authored and
taronaeo
committed
rpc : check for null buffers in get/set/copy tensor endpoints (#14868)

struct
authored and
taronaeo
committed
mtmd : fix 32-bit narrowing issue in export-lora and mtmd clip (#14503)

kiwi142857
authored and
taronaeo
committed
context : restore preemptive sched reset when LLAMA_SET_ROWS=0 (#14870)

ggerganov
authored and
taronaeo
committed
ggml : remove invalid portPos specifiers from dot files (#14838)

ORippler
authored and
taronaeo
committed
ggml-cpu: disable ggml-nnpa compile flag by default
taronaeo
committed
docs: update s390x build docs to reflect nnpa disable
taronaeo
committed