facebookresearch schedule_free issues

facebookresearch / schedule_free

Schedule-Free Optimization in PyTorch

Apache License 2.0

1.9k stars 64 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Is there any "adamw" style wrap_schedulefree?

#51 And233 opened 4 days ago
0
Modernization

#49 adefazio closed 2 weeks ago
0
How to place the optimizer in eval mode when storing checkpoints?

#48 lyndonlauder closed 3 weeks ago
1
LR and weight decay joint tuning

#47 EIFY closed 2 weeks ago
4
[FeatureRequest]Add AdEMAMixScheduleFree

#46 sdbds opened 1 month ago
3
Decoupled Weight Decay?

#45 madman404 closed 2 months ago
1
batchnorm update

#44 gattia closed 2 months ago
1
Potentially breaking DDP training

#43 irowberryFS closed 2 months ago
4
Update tensor device on train and eval function

#42 zhulinchng closed 3 months ago
1
Update tensor device on train and eval function

#41 zhulinchng closed 3 months ago
2
8bit optimizers

#40 winglian closed 2 months ago
1
Warmup when restarting

#39 Fr0do closed 4 months ago
1
Where should beta2 bias correction be incorporated?

#38 jondeuce closed 2 weeks ago
2
How to combine this with Mechanic?

#37 Yura52 closed 2 weeks ago
2
Wrapper Version

#36 adefazio closed 4 months ago
7
Add type hints

#35 stecklin closed 4 months ago
1
Missing type hints (for CLI usage)

#34 stecklin closed 4 months ago
2
How do I retrive current LR from the optimizer now?

#33 pkpro closed 4 months ago
1
When saving/loading optimizer using state_dict(), a mismatch between state["z"] and param (self.state[p]) will occur.

#32 poutyface closed 5 months ago
4
Guidelines for Learning Rates

#31 MarkWijkhuizen closed 4 months ago
2
Doesnot work with torch.compile

#30 yuanzhi-zhu closed 2 weeks ago
2
Reference AdamW?

#29 neosr-project closed 5 months ago
1
Foreach detection

#28 adefazio closed 5 months ago
0
Regular AdamW scores just as well on MNIST? + Feature request...

#27 drscotthawley closed 5 months ago
2
AttributeError: module 'torch' has no attribute '_foreach_lerp_'

#26 patriksimurka closed 5 months ago
1
Paper release

#25 adefazio closed 5 months ago
0
Compatibility with MuP (Maximal Update Parametrization)

#24 simonguozirui closed 3 months ago
2
Typo in paper?

#23 ad8e closed 5 months ago
3
Weights before saving are different from loaded weights while training with AdamWScheduleFree

#22 talrub closed 6 months ago
4
Can this be tested in kohya_ss

#21 rafstahelin closed 5 months ago
3
Fixing typos here and there

#20 konstmish closed 7 months ago
2
Foreach fix

#19 adefazio closed 7 months ago
0
bug going from 1.2.1 to 1.2.2

#18 MarcoForte closed 7 months ago
2
Fix typo in SGDScheduleFree docstring

#17 boisgera closed 7 months ago
2
Retrieve LR

#16 bhack closed 7 months ago
3
Does saved parameters need adjusted?

#15 FykAikawa closed 7 months ago
4
ERROR: Could not build wheels for schedulefree

#14 talrub closed 2 months ago
1
Delete schedulefree/__pycache__ directory

#13 bryant1410 closed 7 months ago
1
Non-closure Prodigy?

#12 madman404 opened 7 months ago
4
batchnorm and intermittent evals

#11 sniklaus closed 7 months ago
4
Add foreach to optimizers

#10 drhead closed 7 months ago
5
Minimal harness for running CIFAR10 experiments

#9 nelaturuharsha opened 7 months ago
4
Usage with Torch's Autocast and GradScaler

#8 fhahlbohm closed 6 months ago
7
Update README.md

#7 eltociear closed 7 months ago
1
Tracking effective learning rate

#6 drhead closed 7 months ago
2
[BUG]ZeroDivisionError: float division by zero

#5 sdbds closed 7 months ago
6
fix small typo, .val() -> .eval()

#4 mrdbourke closed 7 months ago
3
DeepSpeed support?

#3 catid closed 7 months ago
2
nit: grammar

#2 BasedLukas closed 7 months ago
1
Add `examples` folder, `mnist` example

#1 tfaod closed 7 months ago
0