Reduce dependencies for inference

Question

Reduce dependencies for inference

Closed this issue 2 months ago · 2 comments

🎯 Motivation

Currently, the dependencies required for training and inference are bundled together. However, the number of dependencies required for inference is smaller than those needed for training. Given that more users are likely to only use the inference functionality rather than training, it would be beneficial to minimize the dependencies required for inference.

📝 Description

We should separate the dependencies for training and inference.

This will result in the following benefits for users who only use inference:

Shorter installation time when running pip install -e ..
Fewer dependency conflicts when using other libraries alongside this package.

💡 Expected Outcome

Reduce the number of installed packages when only inference is required.
Ensure that users who only need inference can install the minimum necessary dependencies.
Ensure that users who need training can install the necessary dependencies for training with a dedicated option.

📍 Additional Notes

The following dependencies seem unnecessary for inference and could potentially be removed:

wandb
python-dotenv
scikit-learn
pandas

Answer 1 · 2024-09-15T22:59:42.000Z

I'll handle this issue.

Answer 2 · 2024-09-17T01:00:10.000Z

All dependencies listed above was removed by these PRs:

#17
#26 .

The issue was resolved.