 gaopengzhi
|
c7d410725b
Merge branch 'main' into grad_clip
|
vor 1 Jahr |
|
gaopengzhi
|
e2797abe9b
Add gradient_clipping and gradient_clipping_threshold parameters
|
vor 1 Jahr |
|
gaopengzhi
|
bb7c6c1e33
Support FSDP scenario
|
vor 1 Jahr |
|
gaopengzhi
|
b1d9efd155
Refactor gradient clipping feature
|
vor 1 Jahr |
 Jeremy Howard
|
eca8410b32
Use bf16 parameters in bf16 mixed prec
|
vor 1 Jahr |
|
gaopengzhi
|
04befdef69
Add gradient clipping feature
|
vor 1 Jahr |
 Matthias Reso
|
e8bb7fbabc
Merge remote-tracking branch 'origin/main' into feature/length_based_batch_sampling
|
vor 1 Jahr |
|
Matthias Reso
|
33925f71e6
Add missing amp context if use_fp16 is enabled
|
vor 1 Jahr |
|
Matthias Reso
|
10f9367e56
fix missing labels in datasets
|
vor 1 Jahr |
 hongbo.mo
|
6217635e87
Fix tqdm bar not change length after terminal is resized
|
vor 1 Jahr |
|
Matthias Reso
|
c33ea3cacb
Fix pbar update
|
vor 1 Jahr |
|
hongbo.mo
|
5e910e6a42
Fix typo
|
vor 1 Jahr |
|
hongbo.mo
|
0bc6a07a80
bugfix: update tqdm bar with the fixed gradient_accumulation_steps
|
vor 1 Jahr |
|
Matthias Reso
|
72a9832571
Merge branch 'main' into feature/package_distribution
|
vor 1 Jahr |
|
Matthias Reso
|
ce9501f22c
remove relative imports
|
vor 1 Jahr |
|
Matthias Reso
|
5b58afc754
Fix div by zero if run_validation=False
|
vor 1 Jahr |
|
Matthias Reso
|
cf678b9bf0
Adjust imports to package structure + cleaned up imports
|
vor 1 Jahr |
|
Matthias Reso
|
4c9cc7d223
Move modules into separate src folder
|
vor 1 Jahr |
