Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Feature] DoReMi #54

Merged
merged 88 commits into from
Feb 22, 2024
Merged

[Feature] DoReMi #54

merged 88 commits into from
Feb 22, 2024

Conversation

xrsrke
Copy link
Member

@xrsrke xrsrke commented Feb 1, 2024 •
edited
Loading

@xrsrke
migrate doremi
f014ed4
@xrsrke
fix doremi training loop
82da203
@xrsrke
Merge branch 'main' of github.com:huggingface/nanotron into xrsrke/fe…
4efa4ca 
…ature_doremi_new_codebase
@xrsrke
got dataset working
0c54724
@xrsrke
clean up
fc59576
@xrsrke
apply precommit
3b3e66c
@xrsrke
refactor
061ffcf
@xrsrke
refactor
6cd0099
@xrsrke
fuck yea, ComebinedDataset now works with big dataset, thanks god (wh…
a3fbc12 
…oever that is)
@xrsrke
refactor
1458618
@xrsrke
refactor
2d6f6c5
@xrsrke
add reference training
9ae7978
@xrsrke
fixed reference training
adccad1
@xrsrke
add data preprocessing script
ce59fb1
@xrsrke
add loading tokenized dataset from disk
ee48f3c
@xrsrke
add unit tests for dataloader
fddd29e
@xrsrke
fixing sampling bug in doremi sampler
d7ee39e
@xrsrke
add testing synchronization across TP and DP dimensions for DoReMi sa…
8e767df 
…mpler
@xrsrke
add testing stateless doremi sampler
a218e6d
@xrsrke
refactor tests
28df796
@xrsrke
backup
c97772f
@xrsrke
support global sampling
101122b
@xrsrke
i did it. now the DoReMi sampler returns the exact number of samples …
e75cdb4 
…per domain in the global batch size
@xrsrke
update dataloader
262fa40
@xrsrke
referecne training works
d17ecc4
@xrsrke
fix repeating indices in the DoReMi sampler
7a9d3f3
@xrsrke
fix skip dp samples
6e28f9b
@xrsrke
fix yielding too many idxs
c1f2916
@xrsrke
fix missing samples
fa44f63
@xrsrke
revert back to a less dump state
16450b0
@xrsrke xrsrke marked this pull request as ready for review February 13, 2024 09:04
NouamaneTazi
NouamaneTazi requested changes Feb 19, 2024
View reviewed changes
Copy link
Member

@NouamaneTazi NouamaneTazi left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Amazing job 🚀🚀🚀 Can we move srx/nanotron/doremi folder to examples/doremi just like we did with moe? this way we're sure which features correspond to what example and we keep nanotron source code minimal

examples/doremi/README.md Outdated Show resolved Hide resolved
examples/doremi/config_280m_llama.yaml Outdated Show resolved Hide resolved
examples/doremi/train_doremi.py Show resolved Hide resolved
src/nanotron/doremi/utils.py Outdated Show resolved Hide resolved
src/nanotron/doremi/utils.py Outdated Show resolved Hide resolved
examples/doremi/train_doremi.py Outdated Show resolved Hide resolved
src/nanotron/doremi/trainer.py Outdated Show resolved Hide resolved
src/nanotron/doremi/trainer.py Outdated Show resolved Hide resolved
src/nanotron/doremi/trainer.py Outdated Show resolved Hide resolved
@xrsrke
Copy link
Member Author

xrsrke commented Feb 20, 2024

add codespell:
image

bring back isort:
image

NouamaneTazi
NouamaneTazi reviewed Feb 20, 2024
View reviewed changes
src/nanotron/models/fast/llama.py Outdated Show resolved Hide resolved
src/nanotron/trainer.py Outdated Show resolved Hide resolved
src/nanotron/trainer.py Outdated Show resolved Hide resolved
src/nanotron/trainer.py Outdated Show resolved Hide resolved
.pre-commit-config.yaml Show resolved Hide resolved
tests/test_doremi_sampler.py Outdated Show resolved Hide resolved
tests/test_doremi_sampler.py Outdated Show resolved Hide resolved
tests/test_doremi_sampler.py Outdated Show resolved Hide resolved
NouamaneTazi
NouamaneTazi reviewed Feb 20, 2024
View reviewed changes
Copy link
Member

@NouamaneTazi NouamaneTazi left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Very nice integration with nanotron. Last batch of comments before merging this!

examples/doremi/configs/config_280m_llama.yaml Show resolved Hide resolved
examples/doremi/configs/config_280m_llama.yaml Show resolved Hide resolved
examples/doremi/README.md Outdated Show resolved Hide resolved
examples/doremi/README.md Show resolved Hide resolved
examples/doremi/README.md Show resolved Hide resolved
examples/doremi/doremi/utils.py Show resolved Hide resolved
examples/doremi/doremi/trainer.py Outdated Show resolved Hide resolved
examples/doremi/doremi/trainer.py Outdated Show resolved Hide resolved
examples/doremi/doremi/trainer.py Outdated Show resolved Hide resolved
examples/doremi/doremi/dataloader.py Show resolved Hide resolved
NouamaneTazi
NouamaneTazi approved these changes Feb 22, 2024
View reviewed changes
Copy link
Member

@NouamaneTazi NouamaneTazi left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM 🚀 🚀 🚀 🚀
Some minor comments left

examples/doremi/configs/config_2.8b_llama.yaml Outdated Show resolved Hide resolved
examples/doremi/doremi/doremi_context.py Outdated Show resolved Hide resolved
examples/doremi/doremi/trainer.py Show resolved Hide resolved
examples/doremi/train_reference.py Outdated Show resolved Hide resolved
@xrsrke xrsrke merged commit 9f9af42 into main Feb 22, 2024
2 of 3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@NouamaneTazi NouamaneTazi NouamaneTazi approved these changes

@3outeille 3outeille Awaiting requested review from 3outeille

@thomwolf thomwolf Awaiting requested review from thomwolf

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@xrsrke @NouamaneTazi