chore: bump python version

[dushyantbehl]

Description of the change

Related issue number

How to verify the PR

Was the PR tested

• I have added >=1 unit test(s) for every new method I have added.
• I have ensured all unit tests pass

[github-actions]

Thanks for making a pull request! 😃
One of the maintainers will review and advise on the next steps.

[willmj]

Thanks for the quick fix and testing Dushyant! Apologies for the lack of understanding but could you clarify what prevents us from keeping a minimum version of 3.9 for users who may want to use it?

[dushyantbehl]

Thanks for the quick fix and testing Dushyant! Apologies for the lack of understanding but could you clarify what prevents us from keeping a minimum version of 3.9 for users who may want to use it?

@willmj Anything less than v3.12 has a FIPS security compliance issue which stems from underlying python version...if security is not a concern we can keep multiple versions one internal to the product where we go with the compliant 3.12+ python else we keep it open to 3.9 users like we had already.

[willmj]

That makes sense, thanks! This change seems good to me for now, but let's wait for @anhuong's approval before merging.

[anhuong]

I think we should leave all versions to test unit tests on, we can still support python 3.9-3.12

[anhuong]

Even though our minimum version is set to 3.9, we have been building our image with 3.11. I don’t see an issue with moving this up to 3.12 except for the additional tests we should run to ensure everything works as expected. We can still have our minimum version be 3.9 but just has our automation running with 3.12. Currently our unit tests are the only thing that tests the suite of python version whereas our e2e tests focus solely on testing the image.

[willmj]

Testing results on python 3.12 version:

	3.12 Baseline runtime	3.12 Runtime with flags set	3.12 Baseline MF1 Score	3.12 MF1 score with flags set
Granite 3.1 8b FT	3257	3094	0.590	0.641
Granite 3.1 8b LoRA	1033	1043	0.610	0.605
Llama 3.1 8b FT	3482	2956	0.596	0.621
Llama 3.1 8b LoRA	1091	1092	0.577	0.568
PowerLM 3b QLoRA	1027	630	0.607	0.589

for comparison, this is what we saw on python 3.11

	Baseline runtime	Runtime with flags set	Baseline MF1 Score	MF1 score with flags set
Granite 3.1 8b FT	4476	3520	0.589	0.561
Granite 3.1 8b LoRA	1102	1094	0.599	0.601
Llama 3.1 8b FT	4066	3391	0.596	0.645
Llama 3.1 8b LoRA	1110	991	0.587	0.588

[willmj]

LGTM!

[anhuong]

Updating python to 3.12 for fips compliance, image and library will be built with python 3.12 but support will remain for python 3.9-3.12

[praveenj83]

Thanks everyone for the quick turnaround on this!