ggerganov/llama.cpp

Pull Requests Commits

ggml : do not crash when quantizing q4_x_x with an imatrix (#9192)

slaren committed 1 year ago

Verified 7d787ed9

metal : separate scale and mask from QKT in FA kernel (#9189)

ggerganov committed 1 year ago

Verified 06658ad7

ggml : add SSM Metal kernels (#8546)

ggerganov committed 1 year ago

Verified fc18425b

tests : fix compile warnings for unreachable code (#9185)

ggerganov committed 1 year ago

Verified 879275ac

ci : add VULKAN support to ggml-ci (#9055)

ggerganov committed 1 year ago

Verified 7a3df798

server : update deps (#9183)

ggerganov committed 1 year ago

Verified e5edb210

metal : gemma2 flash attention support (#9159)

slaren committed 1 year ago

Verified 0c41e03c

ggml-ci : try to improve build time (#9160)

slaren committed 1 year ago

Verified f12ceaca

llama : fix time complexity of string replacement (#9163)

jart committed 1 year ago

Verified 436787f1

common: fixed not working find argument --n-gpu-layers-draft (#9175)

GermanAizek committed 1 year ago

Verified 93bc3839

CUDA: fix Gemma 2 numerical issues for FA (#9166)

JohannesGaessler committed 1 year ago

Verified f91fc563

CPU/CUDA: Gemma 2 FlashAttention support (#8542)

JohannesGaessler committed 1 year ago

Verified e11bd856

quantize : fix typo in usage help of `quantize.cpp` (#9145)

joaodinissf committed 1 year ago

Verified 8f824ffe

lora : fix llama conversion script with ROPE_FREQS (#9117)

ngxson committed 1 year ago

Verified 3ba780e2

llama : use F32 precision in GLM4 attention and no FA (#9130)

piDack committed 1 year ago

Verified a07c32ea

[SYCL] Add a space to supress a cmake warning (#9133)

qnixsynapse committed 1 year ago

Verified 11b84eb4

[SYCL] Add oneDNN primitive support (#9091)

luoyu-intel committed 1 year ago

Verified 1731d423

llama : simplify Mamba with advanced batch splits (#8526)

compilade committed 1 year ago

Verified a1631e53

server : support reading arguments from environment variables (#9105)

ngxson committed 1 year ago

Verified fc54ef0d

llama : support for `falcon-mamba` architecture (#9074)

younesbelkada committed 1 year ago

Verified b40eb848

llava : zero-initialize clip_ctx structure fields with aggregate initialization 908)

fairydreaming committed 1 year ago

Verified f63f603c

llama : std::move llm_bigram_bpe from work_queue (#9062)

danbev committed 1 year ago

Verified 8455340b

llava: Add ACC OP for GPU acceleration to the Vulkan backend in the LLAVA CLIP model. (#8984)

cyzero-kim committed 1 year ago

Verified 2f3c1466

[SYCL] fallback mmvq (#9088)

airMeng committed 1 year ago

Verified 50addec9

[SYCL] Fix SYCL `im2col` and `convert` Overflow with Large Dims (#9052)

zhentaoyu committed 1 year ago

Verified 4f8d19ff

tests : add missing comma in grammar integration tests (#9099)

fairydreaming committed 1 year ago

Verified 90db8146

cann: add doc for cann backend (#8867)

wangshuai09 committed 1 year ago

Verified cfac111e

rpc : print error message when failed to connect endpoint (#9042)

rgerganov committed 1 year ago

Verified 1b6ff90f

rpc : prevent crashes on invalid input (#9040)

rgerganov committed 1 year ago

Verified 18eaf29f

flake.lock: Update (#9068)

ggerganov committed 1 year ago

Verified 554b0490

Newer Older