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.

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

Python 3.11 + FA 2.5.0 + Torch 2.3.0 #2898

Merged
merged 77 commits into from
Jan 26, 2024
Merged

Python 3.11 + FA 2.5.0 + Torch 2.3.0 #2898

merged 77 commits into from
Jan 26, 2024

Conversation

KuuCi
Copy link
Contributor

@KuuCi KuuCi commented Jan 23, 2024

This PR implements python 3.11 support for torch nightly using flash attention 2.5.0. The docker image has passed corgi regression tests on llm-foundry as well as has been tested in an interactive using the following set up from llm foundry:
mosaicml/llm-foundry#913

@KuuCi KuuCi requested review from a team as code owners January 23, 2024 20:26
Copy link
Contributor

@j316chuck j316chuck left a comment

Choose a reason for hiding this comment

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

LGTM barring one nit

.github/workflows/docker-configure-build-push.yaml Outdated Show resolved Hide resolved
Copy link
Contributor

@j316chuck j316chuck left a comment

Choose a reason for hiding this comment

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

Thanks! Please add the CORGI run and the a test run with FA2 on PyTorch 2.3.0 ✅

@j316chuck j316chuck marked this pull request as ready for review January 25, 2024 22:02
composer/datasets/utils.py Outdated Show resolved Hide resolved
@j316chuck j316chuck self-requested a review January 26, 2024 20:56
Copy link
Contributor

@mvpatel2000 mvpatel2000 left a comment

Choose a reason for hiding this comment

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

Is this planned on being merged? or no?

docker/Dockerfile Show resolved Hide resolved
docker/Dockerfile Show resolved Hide resolved
docker/Dockerfile Show resolved Hide resolved
docker/generate_build_matrix.py Show resolved Hide resolved
docker/Dockerfile Show resolved Hide resolved
docker/generate_build_matrix.py Show resolved Hide resolved
@KuuCi KuuCi merged commit 1eb61d4 into dev Jan 26, 2024
31 checks passed
@KuuCi KuuCi deleted the space-debugger-vincent branch January 26, 2024 22:40
@KuuCi KuuCi changed the title Space debugger vincent Python 3.11 + FA 2.5.0 + Torch 2.3.0 Jan 31, 2024
ShashankMosaicML pushed a commit to ShashankMosaicML/composer that referenced this pull request Feb 3, 2024
* image version update

* update builder

* remove torch 1.13

* possible snappy fix

* moved changes to generate_build_matrix

* 3.11 support

* test

* version test

* remove snappy test

* add 3.10 + 3.11

* potential snappy fix

* nightly patch

* debug

* debug

* extrapolated pytorch to depend on python version

* python 3.8 deprecation assertion

* removed deprecation

* removing import for test

* lint

* lint

* pr review changes

* apt install snappy before pip install

* lint

* disk usage print logs

* du depth 3

* syntax

* more syntax

* syntax

* syntax

* inspect root

* depth 1

* debug

* remove sys and proc from du

* install fa2 through pip

* install dependancy

* no build isolation

* setuptools

* downgrade to 2.3.6

* 2 workers

* revert

* flash 1.0.9

* flash 2.3.6

* lint

* fa 2.5.0

* nightly 3.11

* type

* remove python 3.11 and torch 2.1.2

* remove timeout

* reset latest version

* smoke test update

* lint

* update yaml

* 2.3.6 test

* revert test

* reversion continued

* restoring from before reversion

* max jobs

* increase timeout

* increase timeout

* revert to only include nightly change

* reset to default build time

* update docker yaml

* new names

* merge

* fix merge

* lint

* cpu-3.11-nightly test

* temp rm test

* cpu unit tst

* rm test
ShashankMosaicML pushed a commit to ShashankMosaicML/composer that referenced this pull request Feb 3, 2024
* image version update

* update builder

* remove torch 1.13

* possible snappy fix

* moved changes to generate_build_matrix

* 3.11 support

* test

* version test

* remove snappy test

* add 3.10 + 3.11

* potential snappy fix

* nightly patch

* debug

* debug

* extrapolated pytorch to depend on python version

* python 3.8 deprecation assertion

* removed deprecation

* removing import for test

* lint

* lint

* pr review changes

* apt install snappy before pip install

* lint

* disk usage print logs

* du depth 3

* syntax

* more syntax

* syntax

* syntax

* inspect root

* depth 1

* debug

* remove sys and proc from du

* install fa2 through pip

* install dependancy

* no build isolation

* setuptools

* downgrade to 2.3.6

* 2 workers

* revert

* flash 1.0.9

* flash 2.3.6

* lint

* fa 2.5.0

* nightly 3.11

* type

* remove python 3.11 and torch 2.1.2

* remove timeout

* reset latest version

* smoke test update

* lint

* update yaml

* 2.3.6 test

* revert test

* reversion continued

* restoring from before reversion

* max jobs

* increase timeout

* increase timeout

* revert to only include nightly change

* reset to default build time

* update docker yaml

* new names

* merge

* fix merge

* lint

* cpu-3.11-nightly test

* temp rm test

* cpu unit tst

* rm test
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants