Code

b3631#309

Code

Merged

Nexesenex merged 11 commits intoNexesenex:skystreamNexesenex/croco.cpp:skystreamfrom ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboardAug 27, 2024

+420-110

Commits

Commits on Aug 25, 2024

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

  JohannesGaessler

  authored
  f91fc56

  Copy full SHA for f91fc56

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

  GermanAizek

  authored
  93bc383

  Copy full SHA for 93bc383

Commits on Aug 26, 2024

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

  Show description for 436787f
  jart

  authored
  436787f

  Copy full SHA for 436787f

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

  slaren

  authored
  f12ceac

  Copy full SHA for f12ceac

• metal : gemma2 flash attention support (#9159)

  slaren

  authored
  0c41e03

  Copy full SHA for 0c41e03

• server : update deps (#9183)

  ggerganov

  authored
  e5edb21

  Copy full SHA for e5edb21

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

  ggerganov

  authored
  7a3df79

  Copy full SHA for 7a3df79

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

  Show description for 879275a
  ggerganov

  authored
  879275a

  Copy full SHA for 879275a

• ggml : add SSM Metal kernels (#8546)

  Show description for fc18425
  ggerganov

  authored
  fc18425

  Copy full SHA for fc18425

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

  Show description for 06658ad
  ggerganov

  authored
  06658ad

  Copy full SHA for 06658ad

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

  slaren

  authored
  7d787ed

  Copy full SHA for 7d787ed