Code

[pull] master from ggerganov:master#36

Code

Closed

pull[bot] wants to merge 12 commits intoteleprint-me:masterteleprint-me/llama.cpp:masterfrom ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboard

+1,522-331

Commits

Commits on Feb 11, 2024

• server : add llama2 chat template (#5425)

  Show description for 907e08c
  

  ngxson

  and

  cebtenzzre

  authored
  907e08c

  Copy full SHA for 907e08c

• lookup: add print for drafting performance (#5450)

  JohannesGaessler

  authored
  e4640d8

  Copy full SHA for e4640d8

• ggml : add mmla kernels for quantized GEMM (#4966)

  Show description for a07d0fe
  snadampal

  authored
  a07d0fe

  Copy full SHA for a07d0fe

• ggml : fix compile warnings (unused vars) (#4966)

  ggerganov

  committed
  0f2411f

  Copy full SHA for 0f2411f

• common : fix compile warning

  ggerganov

  committed
  139b62a

  Copy full SHA for 139b62a

• main : ctrl+C print timing in non-interactive mode (#3873)

  ggerganov

  committed
  85910c5

  Copy full SHA for 85910c5

• server : allow to specify tokens as strings in logit_bias (#5003)

  Show description for 6847801
  

  z80maniac

  and

  ggerganov

  authored
  6847801

  Copy full SHA for 6847801

• common : use enums for sampler types (#5418)

  Show description for a803333
  

  z80maniac

  and

  ggerganov

  authored
  a803333

  Copy full SHA for a803333

• vulkan: only use M-sized matmul on Apple GPUs (#5412)

  Show description for c88c74f
  slp

  authored
  c88c74f

  Copy full SHA for c88c74f

• flake.lock: Update

  Show description for 97a3365
  

  github-actions[bot]

  authored and

  philiptaron

  committed
  97a3365

  Copy full SHA for 97a3365

• Add support for BERT embedding models (#5423)

  Show description for 2891c8a
  

  3 peopleauthored
  2891c8a

  Copy full SHA for 2891c8a

• CUDA: mul_mat_vec_q tiling, refactor mul mat logic (#5434)

  Show description for 3bdc4cd
  

  JohannesGaessler

  and

  slaren

  authored
  3bdc4cd

  Copy full SHA for 3bdc4cd