I wish to draw attention to the lack of support for Intel GPUs by the open-source CTranslate2 [1] library developed under the OpenNMT project.
Currently, CTranslate2 supports NVidia GPUs, but AMD developer relations have now stepped in and will be offering support for adding AMD GPUs [2]. It would be very nice if effort could likewise be dedicated by Intel to add support for Intel GPUs.
This project is used by the largest open-source language translation models (e.g. argos-translate and LibreTranslate) but also faster implementations of OpenAI Whisper such as faster-whisper [3]. Hence adding Intel GPU support to this library will have an impact on the open-source ecosystem.
Additionally, the undertaking might serve well as a practical case study in migrating CUDA to C++ with SYCL.
[1] https://github.com/OpenNMT/CTranslate2
[2] https://github.com/OpenNMT/CTranslate2/issues/1072#issuecomment-2012256408
[3] https://github.com/SYSTRAN/faster-whisper
