Burhan

Seems like https://github.com/ultralytics/ultralytics/blob/1503aab0316d345e9d7f8a32c93735ec46c94be3/ultralytics/engine/predictor.py#L93-L94 should be addressing this, but it's not clear as to where `model.args.conf` is getting set to `None` _after_ inference is performed. After some additional testing, I found...

Latest test failed due to CUDA OOM error ``` FAILED tests/test_cuda.py::test_predict_sam - torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacity of 79.14...

Test for [Ultralytics CI / Tests (windows-latest, 3.12, latest)](https://github.com/ultralytics/ultralytics/actions/runs/14859090141/job/41719260071?pr=20514) failing appears to be independent of code changes. > Internal server error occurred while resolving "actions/checkout@v4". Internal server error occurred while...

Appreciate the additional commits, I think my search was too narrow! 🚀

@smohan123 thanks! I did open a PR for the generation of the API/SDK https://github.com/HumanSignal/label-studio-client-generator/pull/64 but it only reflects my _best guess_ of what's needed to make it work.

Target for ~~`ultralytics 8.2.0`~~ bumped to `ultralytics 8.3.0` in favor of integrating YOLO-World #8039

@arad2022 you can track the anticipated Target Date on the [YOLO GitHub Project](https://github.com/orgs/ultralytics/projects/14) and this issue can be used to reference any relevant PRs.

All locations found referencing `label_smoothing` directly via GitHub code search. 

Target for `ultralytics 8.3.0`

@mzhadigerov if you look at the "Projects" section above, you'll see you can expand for additional details, including a `Target Date`.