Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 9, 2023

• Merge branch 'master' into feature/int8_try2

  karpathy

  authored
  1f8af82

  Copy full SHA for 1f8af82

Commits on Sep 21, 2023

• Added support for repeated kv weights

  nickypro

  committed
  2dedad6

  Copy full SHA for 2dedad6

• Add checks/config for tied embedding weights

  nickypro

  committed
  d3c25b1

  Copy full SHA for d3c25b1

Commits on Sep 16, 2023

• updated comment .pt -> .bin

  nickypro

  committed
  ffea287

  Copy full SHA for ffea287

• Added CLI dtype code

  nickypro

  committed
  a61173d

  Copy full SHA for a61173d

• Made default hf export torch.float32

  nickypro

  committed
  19f40a2

  Copy full SHA for 19f40a2

• Changed code so that lm_head and token_embed are tied

  nickypro

  committed
  fc11cc3

  Copy full SHA for fc11cc3

• add option to set dtype for export

  nickypro

  committed
  f38055d

  Copy full SHA for f38055d

Commits on Sep 12, 2023

• Added error handling for LlamaConfig import

  nickypro

  committed
  bf9a116

  Copy full SHA for bf9a116

• fixed whitespace

  nickypro

  committed
  6360a53

  Copy full SHA for 6360a53

• added option to export to huggingface format

  nickypro

  committed
  c568f69

  Copy full SHA for c568f69

Commits on Aug 27, 2023

• properly handle token embeddings & shared classifier wcls

  atamurad

  committed
  6e52df9

  Copy full SHA for 6e52df9

• draft refactor to use QuantizedTensor in function arguments

  atamurad

  committed
  f850a97

  Copy full SHA for f850a97

Commits on Aug 26, 2023

• draft of int8 attempt number two

  karpathy

  committed
  df80471

  Copy full SHA for df80471

• Get vocab_size from token embeddings size

  janimo

  committed
  2c2b284

  Copy full SHA for 2c2b284

Commits on Aug 21, 2023

• Added load_meta_model() to export.py

  atamurad

  committed
  de00547

  Copy full SHA for de00547

• delete the save_torchscript export file, but copy its content to the new export.py for the future maybe

  karpathy

  committed
  dd61b13

  Copy full SHA for dd61b13

• removed transformers from requirements.txt, added error message

  atamurad

  committed
  0dd8215

  Copy full SHA for 0dd8215

• Fix WQ and WK permutation in huggingface models

  atamurad

  committed
  155475a

  Copy full SHA for 155475a

Commits on Aug 20, 2023

• Added huggingface model loader to export.py

  atamurad

  committed
  09db52c

  Copy full SHA for 09db52c

• use out_file.tell() instead of nbytes += arithmetic

  karpathy

  committed
  f3db92a

  Copy full SHA for f3db92a

Commits on Aug 19, 2023

• make version 1 be the legacy export but with new header. version 2 will be Q8_0 export

  karpathy

  committed
  4df5e2e

  Copy full SHA for 4df5e2e

• oops fix double indent on quantize def

  karpathy

  committed
  4212bd6

  Copy full SHA for 4212bd6

• new model export: versions 0 (legacy) and 1

  karpathy

  committed
  7f551db

  Copy full SHA for 7f551db

Loading