Code

ggml backends interface v1#547

Code

Merged

slaren merged 23 commits intomasterggerganov/ggml:masterfrom ggml-backendggerganov/ggml:ggml-backendCopy head branch name to clipboardOct 6, 2023

+1,414-259

Commits

Commits on Oct 4, 2023

• ggml backends interface v1

  slaren

  committed
  883f0bc

  Copy full SHA for 883f0bc

• move get_alignment from buffer to backend

  slaren

  committed
  c05714f

  Copy full SHA for c05714f

• ggml-cuda : fix ggml_cuda_op_mul_mat_vec_q

  slaren

  committed
  da82697

  Copy full SHA for da82697

• gpt-2 : better check for CPU backend when settings n_threads

  slaren

  committed
  3cf87a3

  Copy full SHA for 3cf87a3

• .gitignore : add .clangd

  slaren

  committed
  319b4bc

  Copy full SHA for 319b4bc

• merge master

  slaren

  committed
  b527d48

  Copy full SHA for b527d48

• restore tests/CMakeLists.txt

  slaren

  committed
  abf2669

  Copy full SHA for abf2669

Commits on Oct 5, 2023

• ggml-backend : code style suggestions (#551)

  Show description for a1fd06c
  

  ggerganov

  and

  slaren

  authored
  a1fd06c

  Copy full SHA for a1fd06c

• gpt-2 : add comments about KV allocation

  ggerganov

  committed
  c4cd2d7

  Copy full SHA for c4cd2d7

• add ggml_backend_is_cpu

  slaren

  committed
  d8b3efc

  Copy full SHA for d8b3efc

• add backend check to ggml_backend_cpu_set_n_threads

  slaren

  committed
  3dbc43a

  Copy full SHA for 3dbc43a

• backend cpu: fix buffer alignment

  slaren

  committed
  b74ffd5

  Copy full SHA for b74ffd5

• fix CUDA_ARCHITECTURES for mmq

  slaren

  committed
  b4ec978

  Copy full SHA for b4ec978

• ggml-alloc : better handle view initialization

  slaren

  committed
  25ce18a

  Copy full SHA for 25ce18a

• ggml-cuda : fix padding clearing

  slaren

  committed
  b42e19c

  Copy full SHA for b42e19c

Commits on Oct 6, 2023

• ggml-backend : metal (#552)

  Show description for 94b0529
  ggerganov

  authored
  94b0529

  Copy full SHA for 94b0529

• gpt-2 : take advantage of Metal unified memory

  ggerganov

  committed
  ce797df

  Copy full SHA for ce797df

• gpt-2 : remove TODO + update comment

  ggerganov

  committed
  e8bc940

  Copy full SHA for e8bc940

• ggml-backend : fix ggml_backend_is_xxx() interface

  ggerganov

  committed
  5ca14ce

  Copy full SHA for 5ca14ce

• gpt-2 : fix build

  ggerganov

  committed
  b22916c

  Copy full SHA for b22916c

• ggml-cuda : cleanup, fix case for src1 not contiguous

  slaren

  committed
  01710cc

  Copy full SHA for 01710cc

• remove commented code

  slaren

  committed
  9cb2626

  Copy full SHA for 9cb2626

• rename ggml_backend_metal_set_n_threads to n_cb

  slaren

  committed
  1ad7c5e

  Copy full SHA for 1ad7c5e