Add XLSum evaluation / unify eval script#12

Open

haileyschoelkopf wants to merge 166 commits intobigscience-workshop:sentence_retrieval_evalfrom haileyschoelkopf:eval_xlsum

+3,602-329

Commits on Apr 3, 2022

Merge pull request bigscience-workshop#8 from bigscience-workshop/sentence_retrieval_eval
vnikouliNLE
authored

Commits on Apr 22, 2022

added script to train tokenizer only on a subset of the dataset
vnikouliNLE
committed
added script to train tokenizer only on a subset of the dataset
vnikouliNLE
committed
updated instructions for samples tokenizer
vnikouliNLE
committed
updated training script: added some extra parameters in the running script, and changed the slurm running settings
vnikouliNLE
committed
added overlap-replace parameter, added possibility to save embedding layer (instead of whole model), added early stopping,
vnikouliNLE
committed
Merge branch 'sentence_retrieval_eval' of https://github.com/bigscience-workshop/multilingual-modeling into remotes/origin/sentence_retrieval_eval
vnikouliNLE
committed

Commits on May 5, 2022

Merge remote-tracking branch 'remotes/origin/sentence_retrieval_eval' into ext_exp
vnikouliNLE
committed

Commits on May 6, 2022

Commits on May 9, 2022

add xlsum script (version #1 )
NickSchoelkopf
committed
add unified eval script
NickSchoelkopf
committed
xlsum separate script
NickSchoelkopf
committed
script bugfixes
NickSchoelkopf
committed

Commits on May 11, 2022

Commits on May 12, 2022

add num_classes arg to model init
NickSchoelkopf
committed

Commits on May 27, 2022

Merge branch 'bigscience-workshop:master' into eval_xlsum
haileyschoelkopf
authored

Commits on May 30, 2022

merge
yongzx
committed
exp-001: finetune gpt-2 model with new tokenizer on fr

Zheng Xin Yong
authored and
yongzx
committed
add run_clm.py
yongzx
committed
update run_clm and run_clm_no_tok
yongzx
committed
reduce per_device_{train, eval}_batch_size and increase {gradient, eval}_accumulation_steps
yongzx
committed
update exp-001 exp-002
yongzx
committed
update README
yongzx
committed
Update README.md
yongzx
committed
Update README.md
yongzx
committed
update exp-002 and exp-004
yongzx
committed
requirements.txt
yongzx
committed
update
yongzx
committed
remove experiments folder
yongzx
committed
korean exp-001
yongzx
committed
update
yongzx
committed
upload scripts
yongzx
committed
remove exp-001
yongzx
committed
remove README
yongzx
committed
added scripts to evaluate each layer of pretrained LM (encoder only) for cross lingual sentence retrieval task

vnikouliNLE
authored and
yongzx
committed
fix due to diff tokenization for ted dataset

vnikouliNLE
authored and
yongzx
committed
added scripts for training model with adapters and embedding layer FT

vnikouliNLE
authored and
yongzx
committed
eval_xnli_de
yongzx
committed
update
yongzx
committed
update
yongzx
committed
update
yongzx
committed
update README
yongzx
committed
update
yongzx
committed
update xnli evaluation
yongzx
committed
added script to train tokenizer only on a subset of the dataset

vnikouliNLE
authored and
yongzx
committed
added script to train tokenizer only on a subset of the dataset

vnikouliNLE
authored and
yongzx
committed
updated instructions for samples tokenizer

vnikouliNLE
authored and
yongzx
committed
updated training script: added some extra parameters in the running script, and changed the slurm running settings

vnikouliNLE
authored and
yongzx
committed
added overlap-replace parameter, added possibility to save embedding layer (instead of whole model), added early stopping,

vnikouliNLE
authored and
yongzx
committed
update eval
yongzx
committed
update eval
yongzx
committed
clean
yongzx
committed
update madx_run_clm
yongzx
committed
adapted xnli script to properly load wte, wpe and adapters

vnikouliNLE
authored and
yongzx
committed
updated the way we save the model; added fp16 training

vnikouliNLE
authored and
yongzx
committed
change zero_shot to cross_lingual
yongzx
committed
load language adapters during inference setting
yongzx
committed
updated tokenizer training script

vnikouliNLE
authored and
yongzx
committed
added xnli zero shot training and eval scripts

vnikouliNLE
authored and
yongzx
committed
added xnli zero shot training and eval scripts

vnikouliNLE
authored and
yongzx
committed
merged with current version

vnikouliNLE
authored and
yongzx
committed
added script to get stats about different tokenizers

vnikouliNLE
authored and
yongzx
committed
added script to get stats about different tokenizers

vnikouliNLE
authored and
yongzx
committed
added script to get stats about different tokenizers

vnikouliNLE
authored and
yongzx
committed
added script to get stats about different tokenizers

vnikouliNLE
authored and
yongzx
committed
fixed tokenizer training with unk token

vnikouliNLE
authored and
yongzx
committed
rename pretrained_model to adapted_model
yongzx
committed
use updated eval_xnli/adapters_xnli_de_vn.py
yongzx
committed
update XNLI
yongzx
committed
Merge branch 'sentence_retrieval_eval' of https://github.com/bigscience-workshop/multilingual-modeling into sentence_retrieval_eval
yongzx
committed

Commits on May 31, 2022

update
yongzx
committed

Commits on Jun 1, 2022

Commits on Jun 2, 2022

Commits on Jun 9, 2022

updated scripts for sentence retrieval eval
vnikouliNLE
committed

Commits on Jun 13, 2022

Commits on Jun 15, 2022

Debugging XNLI
yongzx
committed

Commits on Jun 16, 2022

seems like embedding+adapter wasn't working as expected in the previous version
vnikouliNLE
committed

Commits on Jun 20, 2022

Merge pull request bigscience-workshop#13 from bigscience-workshop/backcompatibility
yongzx
authored

Commits on Jun 21, 2022

Commits on Jun 23, 2022

add comment about tokenizer training
yongzx
committed
use register hook to update subelements
yongzx
committed
don't use mask to zero out grad
haileyschoelkopf
committed

Commits on Jun 24, 2022

Merge pull request bigscience-workshop#20 from haileyschoelkopf/extend-vocab-hailey
yongzx
authored

Commits on Jun 25, 2022

added argument for enabling finetuning mode
Lintang Sutawika
committed
added logic to freeze all param except bias if bitfit is set to true
Lintang Sutawika
committed
switch order
Lintang Sutawika
committed

Commits on Jun 26, 2022

added script
Lintang Sutawika
committed
minor changes in script
Lintang Sutawika
committed
hardcoded DATA_SAMPLES
Lintang Sutawika
committed
changed lline from freezing base_model to transformer
Lintang Sutawika
committed
overlap-replace support for BLOOM
yongzx
committed
Merge branch 'extend-vocab'
yongzx
committed

Commits on Jun 27, 2022

added script to calculate bias changes, wip
Lintang Sutawika
committed
Merge branch 'master' into bitfit
Lintang Sutawika
committed

Commits on Jun 28, 2022

Commits on Jun 29, 2022

Commits on Jun 30, 2022

WIP
yongzx
committed

Commits on Jul 1, 2022

Commits on Jul 2, 2022

change
Lintang Sutawika
committed
Update README.md
Lintang Sutawika
authored
Update README.md
Lintang Sutawika
authored
fixed logic
Lintang Sutawika
committed

Commits on Jul 5, 2022

Commits on Jul 6, 2022

missing output_dir
yongzx
committed