ggml-org/llama.cpp

Pull Requests Commits

tests : add permuted pad

ggerganov committed 11 hours ago

084b3d8d

cuda : extend GGML_OP_PAD to work with non-cont src0

ggerganov committed 12 hours ago

b7c7c950

spec : remove check rate (#19377)

srogmann committed 12 hours ago

Verified 292f6908

ci : add metal server workflows (#19293)

ggerganov committed 12 hours ago

Verified 81ddc60c

revert : "[Model] Qwen3.5 dense and MoE support (no vision) (#19435)" (#19453)

ggerganov committed 12 hours ago

Verified 972f323e

ggml-virtgpu: add backend documentation (#19354)

kpouget committed 13 hours ago

Verified f5e7734f

cmake : add variable to skip installing tests (#19370)

WhyNotHugo committed 19 hours ago

Verified 1e8924fd

[Model] Qwen3.5 dense and MoE support (no vision) (#19435)

pwilkin committed 1 day ago

Verified 39bf692a

CUDA: Fix non-contig rope (#19338)

ORippler committed 1 day ago

Verified e06088da

rpc : update from common.cpp (#19400)

angt committed 1 day ago

Verified 5fa1c190

server : improve context checkpoint logic (#19408)

ggerganov committed 1 day ago

Verified eb449cdf

llama-quantize : cleanup `--help` output (#19317)

ddh0 committed 1 day ago

Verified 5999b50e

ci : remove server job from webui and move slow test (#19424)

CISC committed 2 days ago

Verified 9a5f5779

ci : use -j param correctly when building with sanitizers (#19411)

ggerganov committed 2 days ago

Verified 96441c95

metal : consolidate bin kernels (#19390)

ggerganov committed 2 days ago

Verified 8872ad21

metal : fix event synchronization in cpy_tensor_async (#19402)

ggerganov committed 2 days ago

Verified 34ba7b5a

model : support Step3.5-Flash (#19283)

forforever73 committed 3 days ago

Verified b8311181

gguf-py : bump sentencepiece version (#19319)

Ahajha committed 3 days ago

Verified 3228e772

ggml-webgpu: JIT compile binary operators and handle binding overlaps (#19310)

abhijitramesh committed 3 days ago

Verified 7fbd36c5

sycl: add F16 support for GGML_OP_CEIL (#19306)

NechamaKrashinski committed 3 days ago

Verified 537eadb1

tests: reduce number of FA test permutations (#19381)

jeffbolznv committed 3 days ago

Verified db6adb3c

common : add common_speculative_is_compat() (#19270)

ggerganov committed 3 days ago

Verified dfde5993

unicode : MSVC regex fix (#19340)

Iemand005 committed 3 days ago

Verified 06bf3796

Kimi-Linear support (backend agnostic + MLA KV cache) (#18755)

ymcki committed 3 days ago

Verified 3688c4f5

vulkan: For coopmat2 FA, use fp16 accumulators for the final result (#19376)

jeffbolznv committed 3 days ago

Verified 1946e46f

vulkan: make FA mask/softcap enables spec constants (#19309)

jeffbolznv committed 3 days ago

Verified f9bd518a

metal : skip loading all-zero mask (#19337)

ggerganov committed 3 days ago

Verified 7fcf1ef4

llama : rename llama-sampling to llama-sampler (#19363)

danbev committed 3 days ago

Verified e696cfc0

cuda : cuda graphs now compare all node params (#19383)

ggerganov committed 3 days ago

Verified 3e216476

metal : adaptive CPU/GPU interleave based on number of nodes (#19369)

ggerganov committed 4 days ago

Verified 22cae832

Older