Skip to content

Conversation

@SanBingYouYong
Copy link

@SanBingYouYong SanBingYouYong commented Sep 13, 2024

Tested a new dockerfile set-up based on readme:install instructions.

Main changes: 11.3 instead of 11.7

#163 : specified numpy version 1.26.4 in requirements_113.txt so that 2.0 packages won't be automatically installed and cause compatibility issues.

From the updated readme:

  • Dockerfile_113_clean is a new tested dockerfile that follows the readme instruction instead of the old Dockerfile, mainly it's based on CUDA 11.3 instead of 11.7. It relies on requirements_113.txt, with the only difference of specifying NumPy version 1.26.4 (tested), so that newest NumPy 2.* packages that has compatibility issues with other dependencies will not be installed automatically.
  • Dockerfile_113_domestic includes tested alternative sources (channel/index) for users without a stable connection to official packages.
    • some credits go to discussions on issues like installing tensorflow-gpu from tsinghua pip mirror.
  • The new dockerfile installs the dependencies into conda environment ernerf, and also you still need to download the required files below
    • some other weight files that will be downloaded at runtime (e.g. from huggingface or torch hub) can be manually downloaded and copy pasted following comments in the domestic dockerfile.

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.

1 participant