Skip to content

Conversation

@jaketae
Copy link
Member

@jaketae jaketae commented Aug 4, 2021

This PR adds two more dependencies (transformers, deepspeed) to requirements.txt that are needed to install this repository.

@thomasw21
Copy link
Member

Small question, can we specify version?

@jaketae
Copy link
Member Author

jaketae commented Aug 5, 2021

@thomasw21 We could, and we probably should. I just wasn't sure which version people are using for their dev environment. Here's what pip list gives me on my local. Does this seem about right? If so, I'll go ahead and specify the versions.

Package            Version   Location
------------------ --------- ------------------------------------------------------
certifi            2021.5.30
charset-normalizer 2.0.4
click              8.0.1
deepspeed          0.4.4
filelock           3.0.12
huggingface-hub    0.0.12
idna               3.2
importlib-metadata 4.6.3
joblib             1.0.1
megatron-lm        1.1.5     /Users/jaketae/Documents/Dev/GitHub/Megatron-DeepSpeed
ninja              1.10.2
numpy              1.21.1
packaging          21.0
pip                21.2.2
protobuf           3.17.3
psutil             5.8.0
pybind11           2.7.1
pyparsing          2.4.7
PyYAML             5.4.1
regex              2021.8.3
requests           2.26.0
sacremoses         0.0.45
setuptools         40.8.0
six                1.16.0
tensorboardX       1.8
tokenizers         0.10.3
torch              1.9.0
tqdm               4.62.0
transformers       4.9.1
typing-extensions  3.10.0.0
urllib3            1.26.6
zipp               3.5.0

Copy link
Contributor

@stas00 stas00 left a comment

Choose a reason for hiding this comment

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

Thank you, @jaketae

With one correction

@stas00
Copy link
Contributor

stas00 commented Aug 5, 2021

I don't think hardcoded versions are a great idea unless you know you really need a specific version. Otherwise you run into conflicts with other projects and necessitate a requirement for a dedicated conda env just for this repo. So my vote is no hardcoded versions unless it's needed. Specifying module>min_version is goodness, but again, only if you know need at least min_version of some module.

In general, when in doubt see how HF transformers does it as generally a lot of thought has been put into that project, so mimicking most things there in general should be a good path.

@jaketae
Copy link
Member Author

jaketae commented Aug 5, 2021

So my vote is no hardcoded versions unless it's needed. Specifying module>min_version is goodness, but again, only if you know need at least min_version of some module.

I don't know on top of my head if there is a specific package version we need (other than the branch of deepspeed). If there is, maybe we can specify the minimum version, but unless that's the case I think we can go versionless for now!

README already specifies this in greater detail.
@jaketae jaketae merged commit f919d0c into main Aug 5, 2021
@jaketae jaketae deleted the update-reqs branch August 5, 2021 15:58
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.

4 participants