ggerganov/llama.cpp

Pull Requests Commits

kv-cache : optimize KQ mask construction (#18842)

ggerganov committed 1 hour ago

Verified 2fbde785

ggml webgpu: support for backend sampling (#18880)

reeselevine committed 15 hours ago

Verified a89002f0

ggml : extend ggml_pool_1d + metal (#16429)

ThoreKoritzius committed 1 day ago

Verified 388ce822

docs : update ops.md for CANN backend (#18654)

hipudding committed 1 day ago

Verified 6ba6a3c7

ggml-blas: hide warnings from included BLAS headers (#18818)

DaAwesomeP committed 1 day ago

Verified 0802d4cf

mtmd : Fix ASR for LFM2.5-Audio-1.5B (#18876)

tdakhran committed 1 day ago

Verified c945aaae

common : implement new jinja template engine (#18462)

ngxson committed 1 day ago

Verified c15395f7

Setting mmap and direct_io to false as default in llama-bench.cpp (#18841)

JTischbein committed 1 day ago

Verified aa1dc377

CANN: Remove unused `ggml_cann_get_device` function (#18625)

rauletorresc committed 1 day ago

Verified 4ea2eaac

CANN: fix an issue where get_env was not fully renamed (#18796)

noemotiovon committed 1 day ago

Verified e20fa27a

CANN: support gated linear attn (#18653)

hipudding committed 1 day ago

Verified baa4ba0a

OpenCL: add SOLVE_TRI op support (#18846)

shaofeiqi committed 1 day ago

Verified 785a7100

cuda : print less debug logs when disabling cuda graphs (#18868)

ggerganov committed 1 day ago

Verified 6e7fc8a1

context : do not reserve scheduler for warmups (#18867)

ggerganov committed 1 day ago

Verified be8e3d95

llama : add adaptive-p sampler (#17927)

ddh0 committed 1 day ago

Verified 13f1e4a9

server: improve slots scheduling for n_cmpl (#18789)

ngxson committed 1 day ago

Verified a04c2b06

context : reserve new scheduler when graph topology changes (#18547)

ggerganov committed 2 days ago

Verified 39173bca

CUDA: fix allignment on register spill for FA (#18815)

JohannesGaessler committed 2 days ago

Verified 5c662d21

ggml-cpu: optimize ggml_vec_dot_bf16 for Power9 (#18837)

shalinib-ibm committed 2 days ago

Verified 8cc0ba95

lora: make sure model keep track of associated adapters (#18490)

ngxson committed 2 days ago

Verified a7e6ddb8

model-loader : support bool array sliding window pattern (#18850)

CISC committed 2 days ago

Verified 2a131801

tests : download models only when running ctest (#18843)

angt committed 2 days ago

Verified ec997b4f

hexagon: support for OP_CPY, host buffers now optional, hvx-utils refactoring and optimizations (#18822)

max-krasnyansky committed 2 days ago

Verified cff777f2

CUDA: Factor out and re-use `block_reduce` function (#18785)

ORippler committed 2 days ago

Verified 36f01324

Restore clip's cb() to its rightful glory - extract common debugging elements in llama (#17914)

pwilkin committed 2 days ago

Verified d98b5481

model : clean up and fix EXAONE-MoE configuration (#18840)

nuxlear committed 2 days ago

Verified 8fb71755

refactor : remove libcurl, use OpenSSL when available (#18828)

angt committed 2 days ago

Verified 516a4ca9

vulkan: Check maxStorageBufferRange in supports_op (#18709)

jeffbolznv committed 3 days ago

Verified 3e4bb296

llama-model: fix unfortunate typo (#18832)

am17an committed 3 days ago

Verified 47f96124

CUDA : fix typo in clang pragma comment [no ci] (#18830)

danbev committed 3 days ago

Verified 01cbdfd7

Older