ggerganov/llama.cpp

Pull Requests Commits

server : fix deadlock when removing a finished download

allozaur committed 7 days ago

8040df84

common : resolve <quant>-<sidecar> download tags and list cached sidecars

allozaur committed 7 days ago

be1f1c51

common : fix HF cache paths on Windows (#29475)

angt committed 7 days ago

Verified 6c7a87f7

webgpu: Handle unaligned writes in ggml_backend_webgpu_buffer_set_tensor (#29471)

jbooth committed 7 days ago

Verified f00a64c1

tests : refactor test-recurrent-state-rollback (#29426)

ggerganov committed 7 days ago

Verified d77dd080

ggml-cpu: enable tiled flash attention for non-vector-multiple head dims on x86 (#29423)

SongXiaoXi committed 7 days ago

Verified 6f767fe9

ci : ignore more vgpr spills in > 256 DQK fattn kernels (#29571)

IMbackK committed 7 days ago

Verified f916130d

vulkan: fuse qwen4exp's SCALE -> SIGMOID -> SCALE -> hc_post chain (#29520)

fxgsell committed 7 days ago

Verified 03a667aa

HIP: fix template skip for DKQ > 256 mfma kernels (#29559)

IMbackK committed 7 days ago

Verified c2a9e160

metal: support left and circular padding in GGML_OP_PAD (#29561)

ServeurpersoCom committed 7 days ago

Verified 4364bf72

context : do not re-reserve the scheduler when toggling causal_attn (#28751)

sihanyu03 committed 7 days ago

Verified ed7ac35e

Enables Windows ARM64 build with MSVC cl.exe (#28362)

sarahwu185 committed 7 days ago

Verified 0c6a6a7c

tests : fix ggml init (#29554)

ggerganov committed 7 days ago

Verified 81ef10ea

vulkan: fix wrong results when a mul_mat reads a slice of a larger cache (#28956)

ServeurpersoCom committed 7 days ago

Verified 52624716

server : allow RANK pooling batch splitting for causal LLM rerankers (ie. Qwen3 and Qwen3-VL) (#28876)

timothywang21 committed 8 days ago

Verified 4da63377

common : avoid side effects around params parsing (#29537)

ggerganov committed 8 days ago

Verified a97cce86

common : make string_split<T> throw on invalid values (#29518)

angt committed 8 days ago

Verified 136887b6

convert : export YaRN scaling parameters for PLaMo-3 (#29528)

tokinasin committed 8 days ago

Verified 9adc7f42

ci : bump ty to 0.0.84 (#29529)

CISC committed 8 days ago

Verified 6fd50a40

jinja : add support for dict builtin (#29477)

CISC committed 8 days ago

Verified 33c923db

opencl: refine bin kernel loading condition (#29503)

lhez committed 8 days ago

Verified c9064dde

sycl: FWHT kernels for block widths above 512 (#29243)

bri-prism committed 8 days ago

Verified c8296709

CUDA: tune fp16 tile FlashAttention configs for head sizes 40-112 (#26289)

animeshsri14 committed 8 days ago

Verified 36d7b083

HIP: Enable fattn-mma kernel on cdna for dkq > 256 for large batch sizes (#28907)

IMbackK committed 8 days ago

Verified 2ebd9ae6

vulkan: fix argsort kernel selection for Adreno (#29469)

0cc4m committed 8 days ago

Verified cea74625

common : throw instead of abort on grammar without llguidance (#29516)

angt committed 8 days ago

Verified da6c28eb

RPC: use RDMA completion channel to not spin (#29440)

am17an committed 8 days ago

Verified d7fb90e8

ci : enable GGML_SCHED_DEBUG_REALLOC=1 for ctest workflows (#29514)

ggerganov committed 8 days ago

Verified 7fb2b082

llama-bench : fix OOB access of hf_file (#29515)

angt committed 8 days ago

Verified 187664b5

hrm : fix layer placement of `z_l_init` weight (#29512)

ggerganov committed 8 days ago

Verified 85ca3b52

Older