INNER CODE UNIT · Python

train_loss_log

OpenMeshLab/MeshXL · engine.py:123

                train_loss_log = {k: v.avg for k, v in loss_break_down_avg.items()}
                train_loss_log["learning_rate"] = curr_lr
                logger.log_scalars(train_loss_log, prefix='train', step=curr_iter)
            
            ### saving checkpoints
            
            if accelerator.is_main_process and (curr_iter + 1) % args.save_every == 0:
                save_checkpoint(
                    args.checkpoint_dir,
                    accelerator.unwrap_model(model),
                    optimizer,
                    curr_epoch,
                    args,
                    best_val_metrics,
                    filename=f"checkpoint_{(curr_iter + 1) // 1000}k.pth",
                )
            
            ### pending and doing evaluations: every xxx after xxx iterations

View source record →