 Matthias Reso
|
bcdc39e32b
Added documentation for th batching strategy
|
пре 1 година |
|
Matthias Reso
|
5bceb44542
Fix fsdp_config.pure_bf16 flag in README
|
пре 1 година |
|
Matthias Reso
|
53fd82355f
Add missing changes
|
пре 1 година |
|
Matthias Reso
|
07bcffbf50
clean up unit tests + add batching test
|
пре 1 година |
|
Matthias Reso
|
4c225c65eb
Fix order of concat vs sampler
|
пре 1 година |
|
Matthias Reso
|
f9756ca79d
Added packing test for samsum
|
пре 1 година |
|
Matthias Reso
|
5a359b7bf2
Fix sampler vs batch_sampler
|
пре 1 година |
|
Matthias Reso
|
fe8122daf1
Adapt alpaca dataset to ConcatDataset
|
пре 1 година |
|
Matthias Reso
|
5da84b2913
Fix usage of dataclass for train_config and fsdp_config
|
пре 1 година |
|
Matthias Reso
|
aa5dee241a
Fix unit test to reflect batch packing
|
пре 1 година |
|
Matthias Reso
|
8620ab8ac2
Fix invalid labels for context in custom dataset/oasst1
|
пре 1 година |
|
Matthias Reso
|
52c417b7d5
Merge branch 'fix/invalidate_label_for_chat' into feature/length_based_batch_sampling
|
пре 1 година |
|
Matthias Reso
|
653a79e3dd
Invalidate context in labels for samsum + grammar
|
пре 1 година |
|
Matthias Reso
|
d3015b4c80
Remove max_word from alpaca; lets deal tokenizer deal with truncation
|
пре 1 година |
|
Matthias Reso
|
a647955fc8
Make packing/padding a training setting
|
пре 1 година |
|
Matthias Reso
|
eafea7b366
Invalidate labels in dialog dataset to disable loss
|
пре 1 година |
|
Matthias Reso
|
cc8cc0d3c3
fix grammar dataset
|
пре 1 година |
|
Matthias Reso
|
2e4bd2a665
Resize vocab size to fix idx error
|
пре 1 година |
|
Matthias Reso
|
10f9367e56
fix missing labels in datasets
|
пре 1 година |
|
Matthias Reso
|
f2d02a9362
Add unit test for dis sampler
|
пре 1 година |
|
Matthias Reso
|
be63d9ec39
Remove padding in alpaca ds; remove concat in grammar
|
пре 1 година |
|
Matthias Reso
|
ddf58d205d
Added dist length based batch sampler
|
пре 1 година |
|
Matthias Reso
|
ca41c1c697
Adjust tests to len based batch sampling
|
пре 1 година |
|
Matthias Reso
|
97a7871f4b
Fix seed in test
|
пре 1 година |
|
Matthias Reso
|
17209cdabd
Add license to test file
|
пре 1 година |
|
Matthias Reso
|
d5054ecae9
Move sampler test
|
пре 1 година |
|
Matthias Reso
|
63ce4ce7f6
Moved sampler to data submodule
|
пре 1 година |
|
Matthias Reso
|
f620f3589d
Adds length based batch sampler
|
пре 1 година |
|
Matthias Reso
|
8ac44ef3be
Fix vocab size mismatch in inference due to added pad token
|
пре 1 година |
 Geeta Chauhan
|
40b32ba559
Fix tqdm bar not change length after terminal is resized (#201)
|
пре 1 година |
