ggml-org/llama.cpp

Pull Requests Commits

ggml : block interleaving support for Q4_K quantization for x86 AVX2 architecture (#12332)

Srihari-mcw committed 179 days ago

Verified 3d82dbcb

convert : avoid calls to tokenizer.added_tokens_decoder (#12473)

bartowski1182 committed 179 days ago

Verified 732b5fbf

context : clear sets containing encoder output sequence ids before storing new values (#12470)

fairydreaming committed 179 days ago

Verified 568013d0

CUDA: Improve flash decoding kernel GPU occupancy for BS=1 case (#12183)

gaugarg-nv committed 179 days ago

Verified 517b5ddb

vulkan: optimize iq1 coopmat2 dequant functions (#12427)

jeffbolznv committed 179 days ago

Verified a9b59288

Fix visionOS build and add CI (#12415)

guusw committed 180 days ago

Verified 0fd8487b

llama : add support for GPT2, Bloom and CodeShell tied word embeddings (#12456)

CISC committed 180 days ago

Verified 108e53c2

convert : Support chat_template.json (#12460)

CISC committed 180 days ago

Verified a686171e

vulkan: Submit once enough matmul work has been recorded (#12406)

jeffbolznv committed 180 days ago

Verified c446b2ed

opencl: improve profiling (#12442)

lhez committed 180 days ago

Verified d84635b1

graph : normalize Q, K, V shapes + sync cross attention (#12449)

ggerganov committed 180 days ago

Verified 75422e8b

musa: override warp_size of musa device to 32 (#12445)

yeahdongcn committed 180 days ago

Verified bb115d2b

llama : support converting Mistral Small text-only (#12450)

ngxson committed 180 days ago

Verified 29fff308

speculative : fix seg fault in certain cases (#12454)

ggerganov committed 180 days ago

Verified c6af2161

llama : add support for EXAONE tied word embeddings (#12451)

ngxson committed 180 days ago

Verified 99aa304f

context : always use non-causal attention for encoder graphs (#12447)

ggerganov committed 181 days ago

Verified 8551c44d

SYCL: using graphs is configurable by environment variable and compile option (#12371)

lslusarczyk committed 181 days ago

Verified 35cae5ba

server : fix warmup draft cache type (#12446)

ggerganov committed 181 days ago

Verified 810e0af3

cmake : fix PowerPC build (#12241)

mehendarkarprajwal committed 181 days ago

Verified eba92d64

ggml : add SVE support for q6_K_q8_K (#12361)

fj-y-saito committed 181 days ago

Verified d9a14523

Vulkan: Default to 1GB allocations instead of 4GB to avoid fragmentation and driver issues (#12434)

0cc4m committed 181 days ago

Verified fd123cfe

fixed compilation warnings in ggml-sycl (#12424)

lslusarczyk committed 181 days ago

Verified a53f7f7b

llama: Add support for RWKV v7 architecture (#12412)

MollySophia committed 181 days ago

Verified 7dfad387

docs : bring llama-cli conversation/template docs up-to-date (#12426)

CISC committed 181 days ago

Verified 60c90292

cuda : enable CUDA Graph on CUDA Toolkit < 12.x (#12394)

gaugarg-nv committed 181 days ago

Verified b1b132ef

ggml-vulkan: remove unused find_program(glslc) (#12416)

guusw committed 181 days ago

Verified 01e8f213

vulkan: Add N/2 and N/4 optimized paths in coopmat2 shader (#12312)

jeffbolznv committed 181 days ago

Verified 484a8ab5

vulkan: subgroup size tuning (#12087)

daniandtheweb committed 182 days ago

Verified cf2270e4

vulkan: use fp32 in coopmat2 q4_k dequant function (#12309)

jeffbolznv committed 182 days ago

Verified f07690c9

vulkan: Pad N dimension of B matrix for coopmat2 perf, to avoid bounds checking (#12273)

jeffbolznv committed 182 days ago

Verified 891c6395

Newer Older