[Question]: Does starting with Docker Composer. yml always require re downloading the PyTorch environment?

[yaoq827]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

version v0.22.1
Starting with Docker Composer. yml, the PyTorch environment will be re downloaded every time. How to configure it so that it will not be downloaded again after being downloaded.

[dosubot]

No, starting with Docker Compose does not always require re-downloading the PyTorch environment—but in the current RAGFlow v0.22.1 setup, PyTorch is installed at runtime if it isn't already present in the image. The function responsible for this is pip_install_torch, which runs every time the container starts (unless DEVICE=cpu), causing PyTorch to be downloaded and installed each time the container is recreated or started fresh (source).

To avoid this, you should build a custom Docker image that includes PyTorch during the build process. You can do this by adding PyTorch to your pyproject.toml and uv.lock files, then building the image so that PyTorch is installed at build time, not at runtime. This way, the image already contains PyTorch, and the runtime installation step becomes a no-op (source).

If you use the default prebuilt image and do not modify it, PyTorch will continue to be re-downloaded on every new container start. Building your own image with PyTorch pre-installed is the best way to prevent repeated downloads.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[Magicbook1108]

As you are running on gpu, I guess the answer is Yes.