Left a few comments mostly around:

• Activating conda environment
• ipykernel as dev dependency
• Why is numpy installed separately?

Furthermore, I don't seepyproject.toml anduv.lock files. The reproducibility aspect of UV is totally missed, this PR only uses UV as a pip replacement. Ideally, we want to use UV as a complete package manager.

The reasons why we use AI Tool's Conda environments alongside uv:

1. Isolation and Reusability
   uv allows us to maintain a local venv environment while also leveraging packages from the Conda environment installed by AI Tool. These two environments remain independent, ensuring that uv run can utilize both local dependencies and Conda-installed packages such as PyTorch and TensorFlow. Additionally, uv sync or uv add does not modify the system environment, achieving both isolation (sample's requirements) and package reuse (AI Tool's ipex/itex).

2. CI Validation of AI Tool + Sample
   The purpose of CI (Continuous Integration) is to verify that AI Tool and its associated samples work correctly. we need to use the environment provided by AI Tool, to validate the samples works well with it.

3. Avoiding Redundant Installations
   Without using the AI Tool environment, every sample would need to install pytorch, tensorflow (which is from AI Tool Conda env) separately. This results in duplicate installations, excessive resource consumption, and significantly increases CI run time.

ipykernel issue is resolved by removing --dev
numpy separately installation also removed.

This PR isn’t just about usinguv as a simple replacement forpip. Instead, it takes advantage of AI Tool’s base image while allowing each sample to create its ownminimal local venv. This ensures that every sample remains independent while still reusing the system environment, ultimately enabling CI to validate that AI Tool + sample work correctly.

Thanks for the review@Ankur-singh , we refactor the code and addedpyproject.toml anduv sync

This PR isn’t just about usinguv as a simple replacement forpip. Instead, it takes advantage of AI Tool’s base image while allowing each sample to create its ownminimal local venv. This ensures that every sample remains independent while still reusing the system environment, ultimately enabling CI to validate that AI Tool + sample work correctly.

This makes sense. The PR looks much better now. Thank you so much for all the work.

There is one major question that needs to be answered. Next oneAPI release for AI tools will not have offline installer. All the packages will either be distributed viaapt-get orpip. This change is quite significant for two reasons:

1. We won't have conda environments to start with.
2. UV is good for python packages but not OS level packages.

We will have to give some thought. I will be happy to hear your ideas.

cc@jimmytwei

Oh, I see! I wasn’t aware that the Conda environment would be removed in the next AI Tool release. If that’s the case, the changes in this PR will be suitable for the current release but could also serve as a reference for the next version.

For each sample, useuv to install the required packages individually, such as PyTorch, TensorFlow, and other dependencies. Similar to this PR, the automatically generatedpyproject.toml should be pushed to the repository. Insample.json, sync the packages frompyproject.toml usinguv sync to create an isolated environment, and useuv run to execute each sample independently.

I'm not sure if all the required packages for each sample can already be installed viaapt orpip. If they can, we can refactor theuv workflow accordingly and remove the Conda-related components from the current AI Tool.

If that’s the case, the changes in this PR will be suitable for the current release but could also serve as a reference for the next version.

Yes, I think we should be doing it incrementally.

I'm not sure if all the required packages for each sample can already be installed viaapt orpip.

I'm not sure about it either, we will have to wait for the next release.@jimmytwei do you have any insights about this?