ggerganov/llama.cpp

Pull Requests Commits

context : move the change to llama_context::encode()

ggerganov committed 170 days ago

Verified 29acf2cf

context : always use non-causal attention for encoder graphs

ggerganov committed 170 days ago

Verified a0554c3c

ggml : add SVE support for q6_K_q8_K (#12361)

fj-y-saito committed 170 days ago

Verified d9a14523

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

0cc4m committed 170 days ago

Verified fd123cfe

fixed compilation warnings in ggml-sycl (#12424)

lslusarczyk committed 171 days ago

Verified a53f7f7b

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

MollySophia committed 171 days ago

Verified 7dfad387

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

CISC committed 171 days ago

Verified 60c90292

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

gaugarg-nv committed 171 days ago

Verified b1b132ef

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

guusw committed 171 days ago

Verified 01e8f213

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

jeffbolznv committed 171 days ago

Verified 484a8ab5

vulkan: subgroup size tuning (#12087)

daniandtheweb committed 171 days ago

Verified cf2270e4

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

jeffbolznv committed 171 days ago

Verified f07690c9

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

jeffbolznv committed 171 days ago

Verified 891c6395

vulkan: Adjust coopmat2 tile sizes and selection heuristic (#12258)

jeffbolznv committed 171 days ago

Verified 2f21123c

cmake : enable building llama.cpp using system libggml (#12321)

ckastner committed 171 days ago

Verified 374101fd

SYCL: set extras only on GGML_TYPE_Q4_0 (#12366)

qnixsynapse committed 172 days ago

Verified b3c9a656

llama : fix OLMo-2-0325-32B-Instruct K-norm size (#12400)

CISC committed 172 days ago

Verified 8ba95dca

context : fix init of n_outputs (#12397)

ggerganov committed 172 days ago

Verified dc079cfd

ci : add --symlinks to xcframework zip command (#12409)

danbev committed 172 days ago

Verified 7b61bcc8

llama-tts : add '-o' option (#12398)

marcoStocchi committed 173 days ago

Verified f4c3dd5d

SYCL: Delete redundant plus sign and space (#12391)

aubreyli committed 173 days ago

Verified 3d35d87b

SYCL : support non-contiguous tensors in binary ops (add, sub, etc) (#12399)

fairydreaming committed 173 days ago

Verified b19bd064

[CANN]MUL_MAT optimization (#12382)

noemotiovon committed 174 days ago

Verified 92a39132

Add CLI arg to llama-run to adjust the number of threads used (#12370)

ericcurtin committed 174 days ago

Verified 9f2250ba

main : add -sysf / --system-prompt-file (#12249) (#12250)

CISC committed 174 days ago

Verified 774973b8

Load all MoE experts during warmup (#11571)

fairydreaming committed 174 days ago

Verified 8fcb5636

server: fix "--grammar-file" parameter (#12285)

dodekapod committed 174 days ago

Verified add2a3aa

graph : simplify attn input build for unified KV cache (#12381)

ggerganov committed 174 days ago

Verified c522ce41

hparams : add SWA rope parameters (#12374)

ggerganov committed 174 days ago

Verified 081bee8c

llama : fix Gemma3 SWA KV cache shift (#12373)

ggerganov committed 175 days ago

Verified 84d54755

Older