Project Graphyte - by CodeZero in partnership with Amazon Web Services (AWS) & HP Z Workstations & Solutions. A molecular prediction model build using a GNN backbone for aqueous solubility using MITs research dataset MoleculeNet which is widely recognised for its credibility in scientific research.
As always we built the initial repo on a local Z book and trained the MVP model on an AMD SoC CPU before migrating the project to AWS (Code Commit and S3). The project was then deployed on a p4d instance with 8x NVIDIA A100s, 96x CPU cores (most likely an AMD EPYC 9654), 1TB of memory, NVMe for scratch, EBS for persistent storage and EFS for our dataset.
We always say combining IT excellence, human excellence and Machine Learning excellence is the way to break new ground in innovation.
Thanks to our partnerships for Immersion Cooling with the major oil companies, we have access to circa 25x PhD Chemists and an untapped amount of structured (some unstructured) molecular data.
OUR PARTNERS:
