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

Modular PEFT Objectives, Support for Continued Training, Configurable SavingStragegy #40

Open
wants to merge 64 commits into
base: master
Choose a base branch
from
Open

Modular PEFT Objectives, Support for Continued Training, Configurable SavingStragegy #40

wants to merge 64 commits into from

Conversation

stefanik12
Copy link
Collaborator

This PR incorporates the implementation of modularization for different objectives, allowing concurrent or separate optimisation of multi-PEFT-module language models. This comes with the support of continued training for both traditional, multi-head, and multi-module architectures and corresponding, configurable SavingStrategy, enabling efficient storage of multi-module architectures with a shared base model (traditional PEFT case).

stefanik12 added 30 commits March 26, 2024 15:23
@stefanik12
Support for .gz files loading
8736c71
@stefanik12
Modules merging in LangModule: no_merge_keys_containing param in _par…
db96268 
…tially_merge_models
@stefanik12
Objective.get_dataset: automatic resolution of device and num of eval…
eb1f952 
… samples
@stefanik12
Data loading: support for arbitrary prefix of data files
7504fb0
@stefanik12
Peft-compatible support for continued training
c659dea
@stefanik12
Remove constraints on seq2seq objective & propagation of kwargs to ev…
aeb75e5 
…aluators generate()
@stefanik12
Fix: avoid using offset in eval data iteration
a81d463
@stefanik12
AdaptationArgs.SavingStrategy
f45c009
@stefanik12
Updated checks for generative evaluators
2eacbbc
@stefanik12
Conditional import of Prism base
f9253cb
@stefanik12
Data iteration in parallel mode
475bc30
@stefanik12
Logging of model reload in continued training
dae2584
@stefanik12
Default prefetching of data
a873644
@stefanik12
Checks for language_ids
4c272d3
@stefanik12
Correction in reloading PEFT model within continued training
37d3e8c
@stefanik12
PEFT modules resolution, propagation of peft_config on Objective init
53979f2
@stefanik12
PEFT modules resolution: corrected warning
c7f35e2
@stefanik12
Tokenizer resolution: logging
fab201e
@stefanik12
Corrected peft version
5becfe5
@stefanik12
Persistence of eval objectives
ea868fa
@stefanik12
Eval objectives persistence: fix
8436c00
@stefanik12
Data iteration offset in cont training: logging
1778386
@stefanik12
Multinode data sampling: fix
d5f8690
@stefanik12
CI & unit test for continued training
974a75c
@stefanik12
continued training test use PEFT objectives to save space
c6f9fd4
@stefanik12
Preloaded module support: parametrized merging, logging
1c6e047
@stefanik12
Support for optional saving of objectives' modules
864a65d
@stefanik12
Support for optional saving of objectives' modules
b908555
@stefanik12
Support for optional saving of objectives' modules
7d6800b
@stefanik12
Support for .gz files loading
4f421cc
@stefanik12
AdaptationArgs.SavingStrategy
5c43af6
@stefanik12
Updated checks for generative evaluators
15f67dc
@stefanik12
Conditional import of Prism base
92d7ef1
@stefanik12
Data iteration in parallel mode
f781932
@stefanik12
Logging of model reload in continued training
c465ee8
@stefanik12
Default prefetching of data
097c906
@stefanik12
Checks for language_ids
c0ef287
@stefanik12
Correction in reloading PEFT model within continued training
9bab69f
@stefanik12
PEFT modules resolution, propagation of peft_config on Objective init
99ab307
@stefanik12
PEFT modules resolution: corrected warning
ea58e23
@stefanik12
Tokenizer resolution: logging
190ce59
@stefanik12
Corrected peft version
a270cba
@stefanik12
Persistence of eval objectives
a0f6447
@stefanik12
Eval objectives persistence: fix
8ba558a
@stefanik12
Data iteration offset in cont training: logging
3a1814a
@stefanik12
Multinode data sampling: fix
437dbc4
@stefanik12
CI & unit test for continued training
50ce96a
@stefanik12
continued training test use PEFT objectives to save space
ef34f7f
@stefanik12
Preloaded module support: parametrized merging, logging
dae7003
@stefanik12
Support for optional saving of objectives' modules
8f4654e
@stefanik12
Support for optional saving of objectives' modules
c828b66
@stefanik12
Merge remote-tracking branch 'origin/modular-objectives' into modular…
bb48d26 
…-objectives
@stefanik12
Dependencies: upper-bound peft version
09f23e5
@stefanik12
Continued training: lang_module refactoring, test coverage
b935e5d
@stefanik12
Flake
94a3100
@stefanik12
Schedule fix: constrain iteration with a given number of epochs, fix …
feeb97b 
…in epochs increment, simplification of data worker assignment
@stefanik12
Copy link
Collaborator Author

stefanik12 commented Jan 13, 2025 •
edited
Loading

TODOs:

  • Incorporate new functionality to README
  • Add an end-to-end test for the README example

@stefanik12 stefanik12 marked this pull request as ready for review January 25, 2025 20:54
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@stefanik12