ggerganov/llama.cpp

Pull Requests Commits

spec : simplify time measurement using common_time_meas

ggerganov committed 19 hours ago

Verified 08f952f8

spec : various improvements ton ngram-map + docs (#19253)

srogmann committed 19 hours ago

Verified b4d05a3d

Remove pipeline cache mutexes (#19195)

nikhilJain17 committed 23 hours ago

Verified 2dc3ce21

Bump cmake max version (needed for Windows on Snapdragon builds) (#19188)

max-krasnyansky committed 1 day ago

Verified 3bc8d2cf

nix: fix allowUnfreePredicate for packages with multiple licenses (#19237)

typedrat committed 1 day ago

Verified 8a98ba45

create test.sh to enhance the parameters for testing, update the guide, rm useless script (#19243)

arthw committed 1 day ago

Verified 2634ed20

nix: fix nix develop .#python-scripts (#19218)

teto committed 2 days ago

Verified 41ea2614

ggml-hexagon: flash-attention and reduce-sum optimizations (#19141)

chraac committed 2 days ago

Verified 89f10baa

quantize: add option --tensor-type-file to llama-quantize (#18572)

EugeoSynthesisThirtyTwo committed 2 days ago

Verified 3dd95914

mtmd: support MiniCPM-o 4.5(vision only) (#19211)

tc-mb committed 3 days ago

Verified ec6c7421

lookup, lookahead: fix crash when n_ctx not specified (#18729)

pestopoppa committed 3 days ago

Verified 14883391

ngram-mod : fix build [no ci] (#19216)

ggerganov committed 3 days ago

Verified 49277958

opencl: add optimized q8_0 mm kernel for adreno (#18871)

shaofeiqi committed 3 days ago

Verified 971facc3

ggerganov committed 3 days ago

d9a2a4bc

cuda : fix compile warnings (whisper/0)

ggerganov committed 3 days ago

dfd6106c

server : wrap around the "id_slot" parameter (#19207)

ggerganov committed 3 days ago

Verified bbada8bf

Correctly fetch q8_1 quantize pipeline in test as needed by 8a3519b (#19194)

sredman committed 3 days ago

Verified 13f3ebfa

spec : add ngram-mod (#19164)

ggerganov committed 3 days ago

Verified dabaa2e7

jinja : add unordered_map include to value.h [no ci] (#19205)

mseri committed 3 days ago

Verified 2e916f99

memory : clarify comments for r_l and s_l tensors [no ci] (#19203)

danbev committed 3 days ago

Verified f3bc9889

tests : add GQA=20 FA test (#19095)

ggerganov committed 3 days ago

Verified c3b87ceb

convert : add missing return statement for GraniteMoeModel (#19202)

danbev committed 3 days ago

Verified 05625031

memory : remove unused tmp_buf (#19199)

danbev committed 3 days ago

Verified 83bcdf72

docs: Add LlamaLib to UI projects (#19181)

amakropoulos committed 3 days ago

Verified b316895f

add tensor type checking as part of cuda graph properties (#19186)

bssrdf committed 3 days ago

Verified ecbf01d4

sycl: implement GGML_UNARY_OP_SOFTPLUS (#19114)

s8322 committed 3 days ago

Verified 1025fd2c

sycl: implement GGML_OP_TRI (#19089)

RachelMantel committed 3 days ago

Verified c7358ddf

Fix typos in SYCL documentation (#19162)

DDXDB committed 4 days ago

Verified d284baf1

ggml-webgpu: improve flastAttention performance by software pipelining (#19151)

ArberSephirotheca committed 4 days ago

Verified bd90fc74

hexagon: enable offloading to Hexagon on Windows on Snapdragon (#19150)

tboinovski1 committed 4 days ago

Verified ce38a4db

Older