bug(build): Python basic image is too large

[gaocegege]

Description

The CPU image built by envd is about 1GB. I am wondering if our LLB works well. It is too large.

Reproduction

Additional Info

---

Message from the maintainers:

Impacted by this bug? Give it a 👍. We prioritise the issues with the most 👍.

[aseaday]

Maybe you mean pytorch image?

[gaocegege]

https://github.com/tensorchord/envd/blob/main/examples/python-basic/build.envd This one.

[gaocegege]

Ref https://github.com/zhuwenxing/envd_demo/actions/runs/3361616233

Here is a reproducible case about the build speed of envd and docker

[popfido]

If it is a development environment image (which is what envd's target) I think it is quite reasonable?

[popfido]

If it is a production environment image I may think it would be a quite large one.

[popfido]

An idea I would like to host in issues is that does envd try to do only as a development environment manager for general AI for science teams? Or it would also taking exporting to production environment as a further option? (I did not check the roadmap at the time I leave this message)

[kemingy]

An idea I would like to host in issues is that does envd try to do only as a development environment manager for general AI for science teams? Or it would also taking exporting to production environment as a further option? (I did not check the roadmap at the time I leave this message)

• It will #157

BTW, ML production images can be huge (up to 10GB).

[kemingy]

Some related issues:

• #1176

[popfido]

An idea I would like to host in issues is that does envd try to do only as a development environment manager for general AI for science teams? Or it would also taking exporting to production environment as a further option? (I did not check the roadmap at the time I leave this message)

• It will feat(lang): Support serving #157

BTW, ML production images can be huge (up to 10GB).

Yeah, it's a case by case problem. But for production images we just need minimal requirements for runtime(when codes for production are stable), which means they should usually be smaller than their corresponding development images.

For development images, we require a high flexibility to meet debugging/developing functionality. Which obviously needs more tool installations and optional configurations.