Sync master with upstream release b6264#217

Merged

Minh141120 merged 24 commits intodevfrom

update-dev-from-master-2025-08-25-00-12

Aug 25, 2025

Commits on Aug 25, 2025

CANN: Optimize RMS_NORM using cache (#15419)

noemotiovon
authored and
qnixsynapse
committed
readme : model : mtdm : lfm2 improvements (#15476)

tdakhran
authored and
qnixsynapse
committed
server : Support multimodal completion and embeddings prompts in JSON format (#15108)

65a
authored and
qnixsynapse
committed
ggml-cpu: Support Q5_0 and Q5_1 on s390x (#15486)

taronaeo
authored and
qnixsynapse
committed
llama : remove KV cache defragmentation logic (#15473)

ggerganov
authored and
qnixsynapse
committed
cuda : add Pad Reflect 1D support (#14659)

authored and
qnixsynapse
committed
ggml: add `conv3d` op (#15182)

rmatif
authored and
qnixsynapse
committed
model : gpt-oss add response_format support (#15494)

aldehir
authored and
qnixsynapse
committed
ggml WebGPU: add support for quantization types (#15440)

reeselevine
authored and
qnixsynapse
committed
test-opt: allow slight inprecision (#15503)

JohannesGaessler
authored and
qnixsynapse
committed
vulkan: optimize mul_mat_id loading row ids into shared memory (#15427)

jeffbolznv
authored and
qnixsynapse
committed
vulkan : support ggml_mean (#15393)

Acly
authored and
qnixsynapse
committed
vulkan.Dockerfile: install vulkan SDK using tarball (#15282)

yeahdongcn
authored and
qnixsynapse
committed
vulkan: Rewrite synchronization to allow some overlap between nodes (#15489)

jeffbolznv
authored and
qnixsynapse
committed
chat : fix debug build assertion in trim function (#15520)

LaffeyNyaa
authored and
qnixsynapse
committed
scripts: fix compare-llama-bench.py (#15521)

JohannesGaessler
authored and
qnixsynapse
committed
model : add support for Seed-OSS (#15490)

authored and
qnixsynapse
committed
vulkan: optimize rms_norm, and allow the work to spread across multiple SMs (#15281)

jeffbolznv
authored and
qnixsynapse
committed
CUDA: fix half2 -> half conversion for HIP (#15529)

JohannesGaessler
authored and
qnixsynapse
committed
vulkan: workaround MoltenVK compile failure in multi_add (#15506)

authored and
qnixsynapse
committed
vulkan: enable Conv2D for Apple after MoltenVK fixed the bug (#15526)

0cc4m
authored and
qnixsynapse
committed
vulkan: Support FA with any multiple of 8 head sizes (#15537)

jeffbolznv
authored and
qnixsynapse
committed
kv-cache : support layer reuse (#15504)

ggerganov
authored and
qnixsynapse
committed
vulkan: apply MUL_MAT_ID subgroup optimization to non-coopmat devices (#15524)

0cc4m
authored and
qnixsynapse
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Sync master with upstream release b6264#217

Sync master with upstream release b6264#217
Minh141120 merged 24 commits intodevfrom
update-dev-from-master-2025-08-25-00-12

Commits on Aug 25, 2025

CANN: Optimize RMS_NORM using cache (#15419)

readme : model : mtdm : lfm2 improvements (#15476)

server : Support multimodal completion and embeddings prompts in JSON format (#15108)

ggml-cpu: Support Q5_0 and Q5_1 on s390x (#15486)

llama : remove KV cache defragmentation logic (#15473)

cuda : add Pad Reflect 1D support (#14659)

ggml: add `conv3d` op (#15182)

model : gpt-oss add response_format support (#15494)

ggml WebGPU: add support for quantization types (#15440)

test-opt: allow slight inprecision (#15503)

vulkan: optimize mul_mat_id loading row ids into shared memory (#15427)

vulkan : support ggml_mean (#15393)

vulkan.Dockerfile: install vulkan SDK using tarball (#15282)

vulkan: Rewrite synchronization to allow some overlap between nodes (#15489)

chat : fix debug build assertion in trim function (#15520)

scripts: fix compare-llama-bench.py (#15521)

model : add support for Seed-OSS (#15490)

vulkan: optimize rms_norm, and allow the work to spread across multiple SMs (#15281)

CUDA: fix half2 -> half conversion for HIP (#15529)

vulkan: workaround MoltenVK compile failure in multi_add (#15506)

vulkan: enable Conv2D for Apple after MoltenVK fixed the bug (#15526)

vulkan: Support FA with any multiple of 8 head sizes (#15537)

kv-cache : support layer reuse (#15504)

vulkan: apply MUL_MAT_ID subgroup optimization to non-coopmat devices (#15524)

Sync master with upstream release b6264#217Minh141120 merged 24 commits intodevjanhq/llama.cpp:devfrom update-dev-from-master-2025-08-25-00-12janhq/llama.cpp:update-dev-from-master-2025-08-25-00-12Copy head branch name to clipboard

Commits

Commits on Aug 25, 2025

Sync master with upstream release b6264#217
Minh141120 merged 24 commits intodevfrom
update-dev-from-master-2025-08-25-00-12