Spaces:
Running
Running
Commit History
handle dtype for embeddings
29db327
add gradient checkpointing
95a8ed2
fix layernorm
7774483
add partition helpers
2856356
remove bias and minor fixes
180ed1e
add standalone modeling file
6197b2f
Merge pull request #27 from tmabraham/fix-forced-bos-token-on-demo
8f484d9
unverified
pcuenca
commited on
fix forced bos token, also applying BART model to 8 samples now
8d4e13c
Merge pull request #26 from tmabraham/generation-training-demo
fc8c230
unverified
demo for generation, including during training from wandb artifact
c48da33
Merge pull request #23 from khalidsaifullaah/main
eb591ff
unverified
pcuenca
commited on
Merge pull request #24 from borisdayma/feat--log-model-frequently
648e404
unverified
feat: use bart-large-cnn
19d68bb
fix: log metadata
99a1ff5
fix: define function before it is used
d449092
fix: correct arg
283adc6
feat: save model frequently
754f876
feat: split script for small and big runs
5e244d0
feat: update test script
3cccb01
feat: bye bye tensorboard
533b494
feat: use bart large
bb3bfa6
YFCC metadata cleaning and encoding script
2c2f570
Merge pull request #22 from borisdayma/feat-axis
a1c047b
unverified
fix: use correct key
b20769d
fix: log correct metrics
3fef9c1
feat: hardcode eval_steps
4c5e5a7
fix: eval_steps belongs to training_args
900136f
feat: eval_steps already exists in TrainingArguments
0a0080b
Merge branch 'main'
3ddf1c5
feat: set default x-axis
97a008e
feat: log everything through wandb
19070ab
Merge pull request #21 from borisdayma/feat-no_decay
b29bab7
unverified
feat: eval less often for faster training
f0a53ac
Merge pull request #20 from borisdayma/eval-interval
635402d
unverified
feat: no decay option
5a3211f
feat: use common wandb shared folder
7aa2f4b
feat: change default for quick tests
71c757b
feat: hardcoded datasets
e8709a6
Merge pull request #19 from pcuenca/main
f8b0895
unverified
Merge pull request #18 from borisdayma/change-bart-large-demo
395641f
unverified
Add eval_interval to evaluate and log every so often.
566d5f2
Pedro Cuenca
commited on
Notebook to encode splitted YFCC100M files.
82fad8c
Pedro Cuenca
commited on
change bart-large-cnn to bart-large in demo folder
5801f13
Ritobrata Ghosh
commited on
Shift tokens in numpy because the built in shift function stalls.
835ea55
Pedro Cuenca
commited on