ggml-org/llama.cpp

Pull Requests Commits

rpc : update from common.cpp (#19400)

angt committed 3 hours ago

Verified 5fa1c190

server : improve context checkpoint logic (#19408)

ggerganov committed 3 hours ago

Verified eb449cdf

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

ddh0 committed 3 hours ago

Verified 5999b50e

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

CISC committed 11 hours ago

Verified 9a5f5779

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

ggerganov committed 12 hours ago

Verified 96441c95

metal : consolidate bin kernels (#19390)

ggerganov committed 1 day ago

Verified 8872ad21

metal : fix event synchronization in cpy_tensor_async (#19402)

ggerganov committed 1 day ago

Verified 34ba7b5a

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

forforever73 committed 1 day ago

Verified b8311181

gguf-py : bump sentencepiece version (#19319)

Ahajha committed 1 day ago

Verified 3228e772

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

abhijitramesh committed 1 day ago

Verified 7fbd36c5

sycl: add F16 support for GGML_OP_CEIL (#19306)

NechamaKrashinski committed 1 day ago

Verified 537eadb1

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

jeffbolznv committed 1 day ago

Verified db6adb3c

common : add common_speculative_is_compat() (#19270)

ggerganov committed 1 day ago

Verified dfde5993

unicode : MSVC regex fix (#19340)

Iemand005 committed 1 day ago

Verified 06bf3796

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

ymcki committed 2 days ago

Verified 3688c4f5

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

jeffbolznv committed 2 days ago

Verified 1946e46f

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

jeffbolznv committed 2 days ago

Verified f9bd518a

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

ggerganov committed 2 days ago

Verified 7fcf1ef4

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

danbev committed 2 days ago

Verified e696cfc0

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

ggerganov committed 2 days ago

Verified 3e216476

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

ggerganov committed 2 days ago

Verified 22cae832

vulkan: Preprocess FA mask to detect all-neg-inf and all-zero. (#19281)

jeffbolznv committed 2 days ago

Verified 449ec2ab

benches : update models + numbers (#19359)

ggerganov committed 2 days ago

Verified 3795cc1e

docker : fix vulkan build (#19352)

CISC committed 3 days ago

Verified b828e18c

vendor : update BoringSSL to 0.20260204.0 (#19333)

angt committed 3 days ago

Verified a4ea7a18

metal : add diag (#19330)

ggerganov committed 3 days ago

Verified 7a4f97d1

vulkan: fix GPU deduplication logic. (#19222)

okuvshynov committed 3 days ago

Verified a498c75a

vulkan: Set k_load_shmem to false when K is too large (#19301)

jeffbolznv committed 3 days ago

Verified 3409ab84

vulkan: fix non-contig rope (#19299)

jeffbolznv committed 3 days ago

Verified c342c3b9

metal : add missing includes (#19348)

will-lms committed 3 days ago

Verified af252d07

Older