Hugging Face
Models
Datasets
Spaces
Community
Docs
Enterprise
Pricing
Log In
Sign Up
Spaces:
Hazem
/
Fac256xc
like
0
Runtime error
App
Files
Files
Community
Fetching metadata from the HF Docker repository...
main
Fac256xc
/
tools
/
train
/
train.py
Commit History
fix(train): overwrite dropout only when specified
89bc9d4
boris
commited on
May 5, 2022
feat: allow eval only
65bb95f
boris
commited on
May 5, 2022
feat(train): allow editing dropout during training
80d791a
boris
commited on
May 3, 2022
feat: support LR offset (#174)
c3e93df
unverified
boris
commited on
Apr 29, 2022
feat(train): allow nesterov momentum
d08bf8d
boris
commited on
Apr 29, 2022
feat: allow weight decay
acc1a4a
boris
commited on
Apr 28, 2022
feat(train): use new HF _do_init api
6b84155
boris
commited on
Apr 19, 2022
fix: apply learning rate offset only when requested
c6263f3
boris
commited on
Apr 16, 2022
feat(train): arg to offset lr for resumed runs
89b4c45
boris
commited on
Apr 16, 2022
fix: allow non-scanned models (#168)
8ae9176
unverified
boris
commited on
Apr 15, 2022
feat: vmap optimizer (#166)
b993d27
unverified
boris
commited on
Apr 11, 2022
feat: scan layers + gradient checkpointing (#161)
07a6f9a
unverified
boris
commited on
Apr 9, 2022
feat: better multi-node support (#158)
728a3c3
unverified
boris
commited on
Apr 8, 2022
feat(model): allow bias (#152)
361a994
unverified
boris
commited on
Mar 30, 2022
feat(train): google-cloud-storage is optional
02b2308
boris
commited on
Mar 29, 2022
feat(train): rename logged config
955dc20
boris
commited on
Mar 29, 2022
feat: implement transformer variants (#144)
542378c
unverified
boris
commited on
Mar 24, 2022
feat(train): log norm and histograms (#143)
b7b619a
unverified
boris
commited on
Mar 19, 2022
feat(data): super conditioning (#141)
7939874
unverified
boris
commited on
Mar 17, 2022
feat: support pod (#139)
803ccbf
unverified
boris
commited on
Mar 14, 2022
fix: no gradient checkpointing for new model
2e02683
boris
commited on
Feb 23, 2022
feat: no gradient checkpointing for params init
b798ed3
boris
commited on
Feb 23, 2022
fix(train): consider schedule offset
bc4734f
boris
commited on
Feb 23, 2022
feat(train): local jax cache
9f5e879
boris
commited on
Feb 23, 2022
feat: add bucket reference to artifact
d368fb6
boris
commited on
Feb 23, 2022
style: lint
d5d442a
boris
commited on
Feb 22, 2022
feat: handle gradient checkpointing
5173ec7
boris
commited on
Feb 22, 2022
feat: load from bucket
1c4e839
boris
commited on
Feb 22, 2022
feat(train): save to bucket
50498e6
boris
commited on
Feb 22, 2022
feat: reduce artifact space + offset step
34cf91c
boris
commited on
Feb 22, 2022
feat: restore weights on CPU
5f954fc
boris
commited on
Feb 21, 2022
feat(train): simplify tokenizer loading
4cb21dd
boris
commited on
Feb 21, 2022
feat(train): use compilation cache
da9367c
boris
commited on
Feb 7, 2022
feat: log num_parameters early
7cfe576
boris
commited on
Feb 2, 2022
feat(train) - handle multiple nodes (#130)
0952927
unverified
boris
commited on
Feb 1, 2022
feat: handle model parallel
1bb3269
boris
commited on
Feb 1, 2022
feat(train): more custom x-axis
5f28cd2
boris
commited on
Jan 31, 2022
fix(train): opt_state_shape for distributed_shampoo
225b6ff
boris
commited on
Jan 29, 2022
feat(train): split artifact into model/state
fa5b058
boris
commited on
Jan 27, 2022
feat(train): another 25% faster
14abe8c
boris
commited on
Jan 26, 2022
feat(train): overhead from 70% to 1% 🥳
2b7f5f1
boris
commited on
Jan 26, 2022
feat(pjit): follow t5x style
7b5868f
boris
commited on
Jan 26, 2022
fix(train): grads spec
00710bc
boris
commited on
Jan 25, 2022
feat(train): improve pjit speed
f254058
boris
commited on
Jan 25, 2022
fix(train): consider correct batch size
b7c7458
boris
commited on
Jan 25, 2022
feat(train): custom start_preconditioning_step
8149924
boris
commited on
Jan 24, 2022
feat(train): handle distributed_shampoo in pjit
032f623
boris
commited on
Jan 24, 2022
feat(train): distributed_shampoo with pjit
cc34d07
boris
commited on
Jan 24, 2022
fix style
f044cb8
boris
commited on
Jan 23, 2022
feat(train): restore opt_state efficiently
1bfc1b5
boris
commited on
Jan 23, 2022
Previous
1
2
Next