Code

Implement full model finetuning#9

Code

Open

OrianeN wants to merge 22 commits intolascivaroma:masterlascivaroma/PaPie:masterfrom OrianeN:ON/finetuneOrianeN/PaPie:ON/finetuneCopy head branch name to clipboard

+683-54

Commits

Commits on Oct 31, 2024

• Implement full model finetuning

  Show description for de465f4
  OrianeN

  committed
  de465f4

  Copy full SHA for de465f4

• SimpleModel load pretrained model refactor + improve tasks params loading

  Show description for 5d8cfb6
  OrianeN

  committed
  5d8cfb6

  Copy full SHA for 5d8cfb6

• Finetuning: implement load+expand pretrained label encoders

  Show description for b6b41dc
  OrianeN

  committed
  b6b41dc

  Copy full SHA for b6b41dc

• LabelEncoder: Fix upper duplicates in register_upper()

  Show description for 21add94
  OrianeN

  committed
  21add94

  Copy full SHA for 21add94

• Finetune: replace option 'expand_labels' by 'labels_mode'

  Show description for 8a1284b
  OrianeN

  committed
  8a1284b

  Copy full SHA for 8a1284b

• Finetune: load tasks with AttentionalDecoder + fix missing sent RNN encoder loading

  OrianeN

  committed
  da1cf3c

  Copy full SHA for da1cf3c

• Enable forcing seed from command-line or config file

  Show description for 0024395
  OrianeN

  committed
  0024395

  Copy full SHA for 0024395

• Finetune: Improve logging when loading the pretrained parameters

  OrianeN

  committed
  294b907

  Copy full SHA for 294b907

• Finetune: Fix loading AttentionalDecoder tasks

  OrianeN

  committed
  5b5cc10

  Copy full SHA for 5b5cc10

• New utility script get_pie_model_params.py to show model info

  Show description for 4a96d0c
  OrianeN

  committed
  4a96d0c

  Copy full SHA for 4a96d0c

• Finetune: fix reducing vocabulary

  Show description for 418c0e5
  OrianeN

  committed
  418c0e5

  Copy full SHA for 418c0e5

• Add new finetune option 'replace_fill'

  Show description for 99e8926
  OrianeN

  committed
  99e8926

  Copy full SHA for 99e8926

• Rewrite LabelEncoder completion of mode 'replace_fill' using attribute freqs + refactor LabelEncoder.expand_vocab

  Show description for 295ef23
  OrianeN

  committed
  295ef23

  Copy full SHA for 295ef23

• train.py : Minor print message change

  OrianeN

  committed
  04ecc5e

  Copy full SHA for 04ecc5e

• Add logs during register_upper

  Show description for 44d295c
  OrianeN

  committed
  44d295c

  Copy full SHA for 44d295c

• Improve logs during register_upper

  OrianeN

  committed
  53ef4e3

  Copy full SHA for 53ef4e3

Commits on Nov 4, 2024

• Fix vocab sizes wrt. reserved tokens

  Show description for 0f5ebc6
  OrianeN

  committed
  0f5ebc6

  Copy full SHA for 0f5ebc6

• register_upper - minor logs improvements

  OrianeN

  committed
  b17f55a

  Copy full SHA for b17f55a

Commits on Nov 14, 2024

• Merge pull request #2 from OrianeN/fix_vocab_reserved

  Show description for 717fa51
  OrianeN

  authored
  717fa51

  Copy full SHA for 717fa51

• Finetune: enable effectively excluding loading of parent weights in the wemb, cemb and task modules

  OrianeN

  committed
  fb9a4cb

  Copy full SHA for fb9a4cb

• Merge pull request #3 from OrianeN/fix_finetune_load

  Show description for 6f3a269
  OrianeN

  authored
  6f3a269

  Copy full SHA for 6f3a269

Commits on Dec 10, 2024

• Update README - Finetuning section

  OrianeN

  committed
  a95153a

  Copy full SHA for a95153a