recovered the docker image

[777lefty]

Description

Checked all the modified parts when migrating evaluation infrastructures, finding out that the docker image was modified, thus causing makefile of data lab to fail. went well This PR fixed this.

Changes

• Changed the docker image in compose.yaml

Testing

Ran locally with qwen3-coder-plus

Checklist

• Tests pass locally
• Code follows project style guidelines
• Documentation updated (if needed)

[tareknaser]

The issue isn't the docker image.
Do you mind checking the model’s performance on the leaderboard? You can see all models fail for the same reason

There’s an issue because the Makefile compiles with -m32 so the task fails on different machines. The Makefile can’t be modified either because it’s marked as protected and the task is meant to make sure it remains unchanged.

[tareknaser]

Also please rebase on top of main branch

[777lefty]

The issue isn't the docker image. Do you mind checking the model’s performance on the leaderboard? You can see all models fail for the same reason

There’s an issue because the Makefile compiles with -m32 so the task fails on different machines. The Makefile can’t be modified either because it’s marked as protected and the task is meant to make sure it remains unchanged.

Hi @tareknaser , I checked all the failure reason before: The failure should have been eliminated, because in my preprocess.sh script, I installed the essential tools needed to compile with -m32 on x64 machines -- that's why it also works on my machine. The reason why I kept '-m32' is because the open-sourced experiment is a little outdated and written on IA32, so I kept it as its original version, and installed gcc-multilib and libc6-dev-i386 to enable -m32. However, from my experience, I noticed that an image with pre-installed gcc environment like gcc:12 or gcc:14 might be conflict with the later installed gcc-multilib and libc6-dev-i386, and I noticed that if we use image like xuafeng/swe-go-python, the conflict vanished. That's how it works on my machine, and I believe it should be the solution.

[xuafeng]

@777lefty Thanks for the explanation. Do you think that we can also remove the make-related files from the protected files?
Then, the model can handle even the docker image is not correct.

[tareknaser]

@777lefty Please add a reference solution as well to verify that the task is solvable

[xuafeng]

@777lefty Can you please solve it as soon as possible? We need to merge this PR it by tomorrow. Thanks.