Training vs Inference Workload Split

Current 60/40 split between model training and inference with future market composition implications