TL;DR: Hugging Face released Transformers v5.20.2, fixing a critical bug affecting input type preservation in multimodal models during generation.
Summary: Hugging Face has released Transformers v5.20.2, a patch addressing an issue where input types could be lost within prepare_inputs_for_generation during the generate function call. This fix is particularly important for multimodal models, ensuring correct input handling. The update also includes other minor bug fixes and improvements.
Why it matters: This patch is crucial for developers working with multimodal AI models in Hugging Face Transformers, preventing potential data type errors and ensuring stable generation. Developers should update their Transformers library to v5.20.2 to avoid these issues, especially when dealing with complex input types.
Source: github_releases