Code

llama : refactor model loading code#2620

Code

Merged

ggerganov merged 29 commits intoggufggerganov/llama.cpp:gguffrom gguf-refactor-loadingggerganov/llama.cpp:gguf-refactor-loadingCopy head branch name to clipboardAug 16, 2023

+2,049-1,994

Commits

Commits on Aug 15, 2023

• llama : style formatting + remove helper methods

  ggerganov

  committed
  a82e3a4

  Copy full SHA for a82e3a4

• llama : fix quantization using gguf tool

  ggerganov

  committed
  66ce19a

  Copy full SHA for 66ce19a

• llama : simplify gguf_file_saver

  ggerganov

  committed
  c9c0b75

  Copy full SHA for c9c0b75

• llama : fix method names

  ggerganov

  committed
  6e29ed5

  Copy full SHA for 6e29ed5

• llama : simplify write_header()

  ggerganov

  committed
  5c85332

  Copy full SHA for 5c85332

• llama : no need to pass full file loader to the file saver

  Show description for 9574f41
  ggerganov

  committed
  9574f41

  Copy full SHA for 9574f41

• llama : gguf_file_saver write I32

  ggerganov

  committed
  da424b6

  Copy full SHA for da424b6

• llama : refactor tensor names (#2622)

  Show description for 2d87c9c
  monatis

  authored
  2d87c9c

  Copy full SHA for 2d87c9c

• gguf : initial write API (not tested yet)

  ggerganov

  committed
  5cb9d9a

  Copy full SHA for 5cb9d9a

• gguf : write to file API (not tested)

  ggerganov

  committed
  85ebfb8

  Copy full SHA for 85ebfb8

• gguf : initial write API ready + example

  ggerganov

  committed
  f6ecd15

  Copy full SHA for f6ecd15

• gguf : fix header write

  ggerganov

  committed
  4463965

  Copy full SHA for 4463965

• gguf : fixes + simplify example + add ggml_nbytes_pad()

  ggerganov

  committed
  c9b2f7f

  Copy full SHA for c9b2f7f

• gguf : minor

  ggerganov

  committed
  35177d7

  Copy full SHA for 35177d7

• llama : replace gguf_file_saver with new gguf write API

  ggerganov

  committed
  4ef5e79

  Copy full SHA for 4ef5e79

• gguf : streaming support when writing files

  ggerganov

  committed
  f7a6aa9

  Copy full SHA for f7a6aa9

• gguf : remove oboslete write methods

  ggerganov

  committed
  1751bd4

  Copy full SHA for 1751bd4

• gguf : remove obosolete gguf_get_arr_xxx API

  ggerganov

  committed
  2906d54

  Copy full SHA for 2906d54

• llama : simplify gguf_file_loader

  ggerganov

  committed
  6c3f824

  Copy full SHA for 6c3f824

• llama : move hparams and vocab from gguf_file_loader to llama_model_loader

  ggerganov

  committed
  a02b809

  Copy full SHA for a02b809

• llama : merge gguf-util.h in llama.cpp

  ggerganov

  committed
  afd135a

  Copy full SHA for afd135a

• llama : reorder definitions in .cpp to match .h

  ggerganov

  committed
  f477fb0

  Copy full SHA for f477fb0

• llama : minor simplifications

  ggerganov

  committed
  23248d7

  Copy full SHA for 23248d7

Commits on Aug 16, 2023

• llama : refactor llama_model_loader (WIP)

  Show description for 5339b85
  ggerganov

  committed
  5339b85

  Copy full SHA for 5339b85

• llama : fix shape prints

  ggerganov

  committed
  31fb56e

  Copy full SHA for 31fb56e

• llama : fix Windows build + fix norm_rms_eps key

  ggerganov

  committed
  c1fe0ab

  Copy full SHA for c1fe0ab

• llama : throw error on missing KV paris in model meta data

  ggerganov

  committed
  f634b29

  Copy full SHA for f634b29

• llama : improve printing + log meta data

  ggerganov

  committed
  e524750

  Copy full SHA for e524750

• llama : switch print order of meta data

  ggerganov

  committed
  6823899

  Copy full SHA for 6823899