dalle-mini / seq2seq /run_seq2seq_flax.py

Commit History

feat: use bart-large-cnn
19d68bb

boris commited on

fix: log metadata
99a1ff5

boris commited on

fix: define function before it is used
d449092

boris commited on

fix: correct arg
283adc6

boris commited on

feat: save model frequently
754f876

boris commited on

feat: use bart large
bb3bfa6

boris commited on

fix: use correct key
b20769d

boris commited on

fix: log correct metrics
3fef9c1

boris commited on

feat: hardcode eval_steps
4c5e5a7

boris commited on

fix: eval_steps belongs to training_args
900136f

boris commited on

feat: eval_steps already exists in TrainingArguments
0a0080b

boris commited on

Merge branch 'main'
3ddf1c5

boris commited on

feat: set default x-axis
97a008e

boris commited on

feat: log everything through wandb
19070ab

boris commited on

Merge pull request #21 from borisdayma/feat-no_decay
b29bab7
unverified

boris commited on

feat: eval less often for faster training
f0a53ac

boris commited on

Merge pull request #20 from borisdayma/eval-interval
635402d
unverified

boris commited on

feat: no decay option
5a3211f

boris commited on

feat: use common wandb shared folder
7aa2f4b

boris commited on

feat: hardcoded datasets
e8709a6

boris commited on

Add eval_interval to evaluate and log every so often.
566d5f2

Pedro Cuenca commited on

Shift tokens in numpy because the built in shift function stalls.
835ea55

Pedro Cuenca commited on

fix: should be converted to array
945d86c

boris commited on

fix: labels array
6c1f112

boris commited on

fix: typo
678a62f

boris commited on

fix: model config
0be4942

boris commited on

fix: correct decoder_input_ids and labels
19946be

boris commited on

feat: don't log model by default
5b79afd

boris commited on

feat: fix typo
ec8d66b

boris commited on

feat: log model
1c44a7d

boris commited on

feat: update default parameters
dbe8c41

boris commited on

feat: output_length considers bos and eos
8bb2236

boris commited on

feat: shared cache folder
42ce7dd

boris commited on

fix: accumulation vs lr
4d55db6

boris commited on

fix: wandb logging with sync_tensorboard
8ba598c

boris commited on

feat: lower default source length
48c07ca

boris commited on

fix: typos
5960e87

boris commited on

feat: gradient accumulation
c9e9575

boris commited on

feat: simplify loss function
9db361a

boris commited on

feat: padding mask not required
d61405b

boris commited on

fix: typo
833a2d5

boris commited on

feat: add adafactor
600ad79

boris commited on

feat: log metrics more frequently
498559f

boris commited on

* Make padding mask optional.
ecafe5e

Pedro Cuenca commited on

* Only perform validation if requested
32dc2d8

Pedro Cuenca commited on

Preprocessing: return "labels", "decoder_input_ids" and
df3c7bd

Pedro Cuenca commited on

Decoder: set eos to an unreachable value, set min_length=max_length to
a841a4c

Pedro Cuenca commited on

accept tsv files as inputs.
a104edb

Pedro Cuenca commited on

val_max_target_length set to OUTPUT_LENGTH
6c27b0d

Pedro Cuenca commited on

use pylogging to refer to logging.
d9f5a35

Pedro Cuenca commited on