Worked on the ecmwf/anemoi-inference repository to improve the reliability of the inference runner, specifically addressing issues with single-variable predictions. Focused on deep learning and inference workflows using Python and PyTorch, the developer identified and resolved a crash caused by unintended dimension collapse during single-variable inference. By carefully adjusting the torch.squeeze operation, the fix preserved the expected 2D output shape, ensuring compatibility with downstream systems and maintaining the integrity of the machine learning pipeline. This targeted bug fix enhanced the stability of the inference process, reducing the risk of errors in production environments and supporting robust model deployment practices.