diff --git a/crates/ml-alpha/examples/alpha_train.rs b/crates/ml-alpha/examples/alpha_train.rs index 79a299ac9..ea594aa1e 100644 --- a/crates/ml-alpha/examples/alpha_train.rs +++ b/crates/ml-alpha/examples/alpha_train.rs @@ -455,6 +455,19 @@ fn main() -> Result<()> { epoch_train_steps += 1; train_loss_running += loss; train_steps += 1; + // Periodic per-step log (every 500 steps) for liveness + + // intra-epoch trajectory monitoring. Cheap (~1 line/min on + // L40S at 8000 steps/epoch / 36s = 220 steps/s ⇒ ~16 lines + // per 500-step interval per epoch). Consumed by + // /tmp/alpha_monitor.py. + if epoch_train_steps % 500 == 0 { + tracing::info!( + epoch, + step = epoch_train_steps, + loss = loss, + "step" + ); + } snap_batch.clear(); label_batch.clear(); }