Kastan | - PDG colab: They need data pipelines for ML & ML training pipelines (see post-meeting email for details). My concern: Codeflare satisfies only a small part of their needs, they primarily need data ops (like Airflow and Spotify's Luigi).
- IN-CORE colab: They have data, need customized ML. I can help them use Codeflare to get better ML performance.
- Ensure Dask worker.distributed for XGBoost works on HAL (currently, workers getting killed by garbage collector)
- Talk to HAL admins about Ray support (for IBM Codeflare). Currently unsupported on
ppcle .
|