Announcement
Atomwell announced as a foundation model for predicting experimental outcomes
The announcement says models of 800 million to 9 billion parameters were trained on roughly 80 million experimental measurements.
TLDR
Atomwell was announced as a foundation model for predicting experimental outcomes. The team says it trained models ranging from 800 million to 9 billion parameters on roughly 80 million experimental measurements and found clean scaling laws for both model size and data in molecular discovery.
Combined views
15.8K
2 Sources, first seen ago
