Code

Merge b3565#286

Code

Merged

Nexesenex merged 61 commits intoNexesenex:patch-1Nexesenex/croco.cpp:patch-1from ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboardAug 10, 2024

+4,783-1,344

Commits

Commits on Aug 2, 2024

• Fix conversion of unnormalized BF16->BF16 weights (#7843)

  Show description for b72c20b
  

  CISC

  and

  compilade

  authored
  b72c20b

  Copy full SHA for b72c20b

Commits on Aug 3, 2024

• ggml : reading the runtime sve config of the cpu (#8709)

  Show description for 76614f3
  

  jdomke

  and

  domke

  authored
  76614f3

  Copy full SHA for 76614f3

Commits on Aug 4, 2024

• flake.lock: Update (#8847)

  ggerganov

  authored
  4b77ea9

  Copy full SHA for 4b77ea9

• baby-llama : remove duplicate vector include

  

  danbev

  authored and

  ggerganov

  committed
  01aae2b

  Copy full SHA for 01aae2b

• batched-bench : handle empty -npl (#8839)

  Show description for ecf6b7f
  

  3 peopleauthored
  ecf6b7f

  Copy full SHA for ecf6b7f

• Server: Don't ignore llama.cpp params (#8754)

  Show description for 978ba3d
  ardfork

  authored
  978ba3d

  Copy full SHA for 978ba3d

• Install curl in runtime layer (#8693)

  bsquizz

  authored
  0d6fb52

  Copy full SHA for 0d6fb52

Commits on Aug 5, 2024

• cann: support q4_0 model (#8822)

  wangshuai09

  authored
  c02b0a8

  Copy full SHA for c02b0a8

• ggml : move c parameter comment to ggml_rope_ext (ggml/901)

  Show description for 655858a
  

  danbev

  authored and

  ggerganov

  committed
  655858a

  Copy full SHA for 655858a

• vulkan : implement Stable Diffusion operators (ggml/904)

  Show description for a3738b2
  

  0cc4m

  authored and

  ggerganov

  committed
  a3738b2

  Copy full SHA for a3738b2

• sync : ggml

  Show description for 5587e57
  ggerganov

  committed
  5587e57

  Copy full SHA for 5587e57

• vulkan : fix Qantized Mat-Vec Mul on AMD GPUs for ncols < 64 (#8855)

  Show description for 064cdc2
  0cc4m

  authored
  064cdc2

  Copy full SHA for 064cdc2

• llama : better replace_all (#8852)

  ggerganov

  authored
  f1ea514

  Copy full SHA for f1ea514

• readme : update model list (#8851)

  BarfingLemurs

  authored
  400ae6f

  Copy full SHA for 400ae6f

• cmake: fix paths for vulkan shaders compilation on Windows (#8573)

  Show description for e31a4f6
  stduhpf

  authored
  e31a4f6

  Copy full SHA for e31a4f6

• Stop the generation when <|eom_id|> token is encountered - needed for Llama 3.1 tool call support (#8858)

  Show description for d3f0c71
  

  fairydreaming

  and

  sszymczy

  authored
  d3f0c71

  Copy full SHA for d3f0c71

• py: Add more authorship metadata from model card (#8810)

  Show description for 1ef14b3
  mofosyne

  authored
  1ef14b3

  Copy full SHA for 1ef14b3

• ggml : fix overflows in elu function (#8866)

  Show description for b9dfc25
  jart

  authored
  b9dfc25

  Copy full SHA for b9dfc25

• readme : add ramalama to the availables UI (#8811)

  Show description for b42978e
  ericcurtin

  authored
  b42978e

  Copy full SHA for b42978e

• cann: fix buffer_num and runtime speed slowly error (#8865)

  wangshuai09

  authored
  bc0f887

  Copy full SHA for bc0f887

• common : Changed tuple to struct (TODO fix) (#8823)

  Show description for 0a4ce78
  Septa2112

  authored
  0a4ce78

  Copy full SHA for 0a4ce78

Commits on Aug 6, 2024

• [SYCL] correct cmd name (#8877)

  arthw

  authored
  d4ff847

  Copy full SHA for d4ff847

• [CANN]: Fix ggml_backend_cann_buffer_get_tensor (#8871)

  Show description for c21a896
  MengqingCao

  authored
  c21a896

  Copy full SHA for c21a896

• convert : add support for XLMRoberta embedding models (#8658)

  Show description for cdd1889
  iamlemec

  authored
  cdd1889

  Copy full SHA for cdd1889

• ggml : add epsilon as a parameter for group_norm (#8818)

  Show description for 2d5dd7b
  MollySophia

  authored
  2d5dd7b

  Copy full SHA for 2d5dd7b

• contributing : add note about write access

  ggerganov

  authored
  0bf16de

  Copy full SHA for 0bf16de

• [Vulkan] Fix compilation of vulkan-shaders-gen on w64devkit after e31a4f6 (#8880)

  Show description for efda90c
  MaggotHATE

  authored
  efda90c

  Copy full SHA for efda90c

• cmake : Link vulkan-shaders-gen with pthreads (#8835)

  Show description for db20f50
  Patater

  authored
  db20f50

  Copy full SHA for db20f50

• simple : update name of executable to llama-simple (#8885)

  Show description for 5f4dcb1
  danbev

  authored
  5f4dcb1

  Copy full SHA for 5f4dcb1

• CUDA: fix padding logic for FP16/FP32 (#8884)

  JohannesGaessler

  authored
  641f5dd

  Copy full SHA for 641f5dd

• server : add lora hotswap endpoint (WIP) (#8857)

  Show description for 1e6f655
  ngxson

  authored
  1e6f655

  Copy full SHA for 1e6f655

• typo correction (#8891)

  Nexesenex

  authored
  3195854

  Copy full SHA for 3195854

• quantize : update usage comment in quantize.cpp (#8889)

  Show description for 725e3d9
  danbev

  authored
  725e3d9

  Copy full SHA for 725e3d9

Commits on Aug 7, 2024

• llama-bench : add support for getting cpu info on Windows (#8824)

  Show description for 506122d
  

  kylo5aby

  and

  slaren

  authored
  506122d

  Copy full SHA for 506122d

• CUDA/HIP: fix tests/test-backend-ops (#8896)

  JohannesGaessler

  authored
  a8dbc6f

  Copy full SHA for a8dbc6f

• [SYCL] Updated SYCL device filtering (#8901)

  Show description for 0478174
  OuadiElfarouki

  authored
  0478174

  Copy full SHA for 0478174

• ggml-backend : fix async copy from CPU (#8897)

  Show description for be55695
  slaren

  authored
  be55695

  Copy full SHA for be55695

• make : use C compiler to build metal embed object (#8899)

  Show description for 15fa07a
  slaren

  authored
  15fa07a

  Copy full SHA for 15fa07a

Commits on Aug 8, 2024

• make : clean llamafile objects (#8923)

  Show description for ebd541a
  DrDub

  authored
  ebd541a

  Copy full SHA for ebd541a

• metal : add abort callback (ggml/905)

  

  conradev

  authored and

  ggerganov

  committed
  85fca8d

  Copy full SHA for 85fca8d

• metal : fix struct name (ggml/912)

  Show description for 5b33ea1
  ggerganov

  committed
  5b33ea1

  Copy full SHA for 5b33ea1

• ggml : ignore more msvc warnings (ggml/906)

  

  iboB

  authored and

  ggerganov

  committed
  f93d49a

  Copy full SHA for f93d49a

• sync : ggml

  ggerganov

  committed
  e44a561

  Copy full SHA for e44a561

• scripts : fix sync filenames (#0)

  ggerganov

  committed
  366d486

  Copy full SHA for 366d486

• scripts : sync cann files (#0)

  ggerganov

  committed
  afd27f0

  Copy full SHA for afd27f0

• gguf-py : simplify support for quant types (#8838)

  Show description for 3a14e00
  compilade

  authored
  3a14e00

  Copy full SHA for 3a14e00

Commits on Aug 9, 2024

• llama : reduce useless copies when saving session (#8916)

  Show description for 345a686
  compilade

  authored
  345a686

  Copy full SHA for 345a686

• server : add one level list nesting for embeddings (#8936)

  gelim

  authored
  daef3ab

  Copy full SHA for daef3ab

• llama : fix typo in llama_tensor_get_type comment [no ci] (#8937)

  danbev

  authored
  6f6496b

  Copy full SHA for 6f6496b

• embedding : add --pooling option to README.md [no ci] (#8934)

  Show description for 5b2c04f
  danbev

  authored
  5b2c04f

  Copy full SHA for 5b2c04f

• whisper : use vulkan as gpu backend when available (whisper/2302)

  Show description for 70c0ea3
  

  mstephenson6

  authored and

  ggerganov

  committed
  70c0ea3

  Copy full SHA for 70c0ea3

• sync : ggml

  ggerganov

  committed
  4305b57

  Copy full SHA for 4305b57

• llava : support MiniCPM-V-2.5 (#7599)

  Show description for 3071c0a
  

  4 peopleauthored
  3071c0a

  Copy full SHA for 3071c0a

• llama : better replace_all (cont) (#8926)

  Show description for 45a55b9
  ggerganov

  authored
  45a55b9

  Copy full SHA for 45a55b9

• make : fix llava obj file race (#8946)

  Show description for 272e3bd
  ggerganov

  authored
  272e3bd

  Copy full SHA for 272e3bd

• llama : add support for lora adapters in T5 model (#8938)

  Show description for 6afd1a9
  

  fairydreaming

  and

  sszymczy

  authored
  6afd1a9

  Copy full SHA for 6afd1a9

• Merge commit from fork

  ggerganov

  authored
  b72942f

  Copy full SHA for b72942f

Commits on Aug 10, 2024

• gguf-py : fix double call to add_architecture() (#8952)

  Show description for 911b437
  tarilabs

  authored
  911b437

  Copy full SHA for 911b437

• Add support for encoder-only T5 models (#8900)

  Show description for 7c3f55c
  

  fairydreaming

  and

  sszymczy

  authored
  7c3f55c

  Copy full SHA for 7c3f55c

• llama : default n_swa for phi-3 (#8931)

  Show description for 7eb2384
  ngxson

  authored
  7eb2384

  Copy full SHA for 7eb2384

• metal : fix uninitialized abort_callback (#8968)

  slaren

  authored
  6e02327

  Copy full SHA for 6e02327