ggml-org/llama.cpp

Pull Requests Commits

glm5-next : drop the q_absorbed cont

am17an committed 8 days ago

1f5161c8

meta : keep the split for views inside one slab of the split axis

am17an committed 8 days ago

933b9f3d

glm5-next : split MLA heads in pairs for -sm tensor

am17an committed 8 days ago

13500544

meta : keep GGML_OP_ARANGE in ggml.h order

am17an committed 8 days ago

c6b348d2

glm5-next : cont q_absorbed for -sm tensor with non-unified KV

am17an committed 8 days ago

0610f855

meta: support GLM-5.3 next sm tensor

am17an committed 8 days ago

54a01649

hexagon: tiled hvx permute (#31446)

marktodorovich committed 8 days ago

Verified 51fe42c5

dflash: copy hidden features via device copies (#31506)

am17an committed 8 days ago

Verified 2cbca389

docs: update s390x build docs (#31546)

taronaeo committed 9 days ago

Verified 29c720ea

mtmd: Pocket tts - support for voices without voice cloning (#31429)

rjtokenring committed 9 days ago

Verified d02b4c0d

model: add Clover support (#31306)

Maphist0 committed 9 days ago

Verified e4b6c8ca

llama : flash_attn_ext_rows - more efficient unified kv for multi-seq (#29510)

am17an committed 9 days ago

Verified 1b6935b2

imatrix : fix --nextn crash (#31522)

sanmai committed 9 days ago

Verified 3f8af15a

context : simplify setters (#29730)

nikwen committed 9 days ago

Verified cd8427d0

readme : fix links in README.md for maintainer PRs (#31487)

maternion committed 9 days ago

Verified 6b00d562

docs: CANN backend supports BF16 on 910B, document GGML_CANN_NO_PINNED (#31249)

li-lizhe committed 9 days ago

Verified 458fcfc3

ggml-zdnn: share buffers with host (#31481)

taronaeo committed 9 days ago

Verified 6ed81205

ggml : use `vm_allocate` instead of `posix_memalign` on all Apple platforms (#31437)

ykhrustalev committed 9 days ago

Verified ea24b952

opencl: add topk_mask (#31441)

lhez committed 9 days ago

Verified 84d9ce0f

vulkan: add tensor-parallel allreduce function (#25051)

pwilkin committed 9 days ago

Verified dcbb0916

vulkan: improve small-n split-K selection on RDNA3 and RDNA4 (#30146)

ravel7524 committed 9 days ago

Verified 07ec7c8a

vulkan: avoid redundant scans when zeroing skipped rows (#31266)

SG-Amadeus committed 9 days ago

Verified cb02250b

sycl : fast ggml_sycl_comm_exchange() on Intel Arc Pro B70s (#30956)

92es committed 9 days ago

Verified 2c93f0b6

sycl: dmmv code cleanup (#31325)

Titaniumtown committed 9 days ago

Verified 5e18eb02

ui: disable model deletion the server would refuse (#31305)

allozaur committed 9 days ago

Verified 5c2ea1ab

ui: align alert dialog styling with other overlay surfaces (#31312)

allozaur committed 9 days ago

Verified 23426d79

ui: keep queued steering messages in their own conversation (#31304)

allozaur committed 9 days ago

Verified 90080ce0

ui: show interrupted state for tool calls that never returned (#31314)

allozaur committed 9 days ago

Verified 64be95b1

ui: don't reload a loading model on select, prefer it in pre-selection (#31326)

allozaur committed 9 days ago

Verified b12b405c

docs : fix Vulkan example command in build.md (#31530)

Septa2112 committed 9 days ago

Verified aac6da80

Older