Code

Model loading#36

Code

Merged

MenuaB merged 10 commits intomainYerevaNN/YNNtitan:mainfrom model_loadingYerevaNN/YNNtitan:model_loadingCopy head branch name to clipboardJan 24, 2025

+289-40

Commits

Commits on Jan 15, 2025

• add llama 125m, 350m and 750m configs

  tigranfah

  committed
  2433a6c

  Copy full SHA for 2433a6c

Commits on Jan 16, 2025

• add llama 125m, 350m and 750m training config files

  tigranfah

  committed
  f8c9c0c

  Copy full SHA for f8c9c0c

• change ngpus to 8 in submitit

  tigranfah

  committed
  8ea8bee

  Copy full SHA for 8ea8bee

Commits on Jan 19, 2025

• use the already existing aim experiment run when resuming the training

  tigranfah

  committed
  bbe27d7

  Copy full SHA for bbe27d7

Commits on Jan 21, 2025

• remove automatically using the experiment hash of the loaded model

  tigranfah

  committed
  8c5d9a7

  Copy full SHA for 8c5d9a7

Commits on Jan 22, 2025

• add command line arguments to submititpt

  tigranfah

  committed
  f2299ac

  Copy full SHA for f2299ac

Commits on Jan 23, 2025

• init the embedding weights with 0.02 std

  tigranfah

  committed
  f0608b1

  Copy full SHA for f0608b1

• fix types of num_decays and decay_steps_perc arguments

  tigranfah

  committed
  c3a9417

  Copy full SHA for c3a9417

• use truncated normal dist for embedding initialization

  tigranfah

  committed
  c9e0a77

  Copy full SHA for c9e0a77

• remove decay_steps argument

  tigranfah

  committed
  0cb487e

  Copy full SHA for 0cb487e