HYPERGAN/EXPERIMENTSNot connected
TRAINING OBSERVATORY

Experiment

Loading

Run id

Raw events ↗
Completed training step—Waiting for run state
Training steps per second—Average of recent updates
Samples per second—Steps/s × global batch
Batch & topology—Device & per-rank split
Samples seen—Updates × global batch
Time training—Wall clock across attempts
Durable checkpoint—Last saved training state
Generator loss—No published value
Discriminator loss—No published value
Best FID—Snapshot quality

Learning curves

Loading history…

1 step / bucket
Accessible data table Latest training values
Latest training values by metric and attempt
Metric Step Value Attempt

Samples & artifacts

Samples are grouped by name (g generated, x real). Each image name shows its most recent version; drag its history slider to step back. The raw generator output behind a grid is a download on that same card; runs without images keep it as a card of numbers. Artifacts are separate from metric measurements.

Model & configuration

Recorded ParticleGAN formulation, network graphs, loss terms, optimizers, and runtime configuration.