train.galvyn.xyz · lab
read-onlySign in

Playground

Pick a model and a training step, then give it a prompt. Slide back to step 0 to see where it started. Learning rate: 1e-3 vs 1e-2: curves and timeline

Learning rate: 1e-3 vs 1e-2

These are base models: they continue text, they don't answer questions. Chat comes with Phase 2's fine-tuning.