defaults: - model: default - computer: v100 - dataset: osv5m - _self_ - exp: ??? model: val_metrics: _target_: metrics.distance_based.HaversineMetrics acc_radiuses: - 1 - 25 - 200 - 750 - 2500 acc_area: [] aux_data: ${aux_data} test_metrics: _target_: metrics.distance_based.HaversineMetrics acc_radiuses: - 1 - 25 - 200 - 750 - 2500 acc_area: ${areas} aux_data: ${aux_data} datamodule: _target_: data.datamodule.ImageDataModule train_dataset: ${dataset.train_dataset} val_dataset: ${dataset.val_dataset} test_dataset: ${dataset.test_dataset} global_batch_size: ${dataset.global_batch_size} num_workers: ${computer.num_workers} num_nodes: ${computer.num_nodes} num_devices: ${computer.devices} val_proportion: 0.1 trainer: _target_: pytorch_lightning.Trainer devices: ${computer.devices} accelerator: ${computer.accelerator} strategy: ${computer.strategy} num_nodes: ${computer.num_nodes} precision: ${computer.precision} max_epochs: ${max_epochs} logger: _target_: pytorch_lightning.loggers.WandbLogger save_dir: ${root_dir} name: ${experiment_name} project: plonk log_model: False offline: False entity: imaginelab checkpoints: _target_: pytorch_lightning.callbacks.ModelCheckpoint dirpath: ${root_dir}/checkpoints/${experiment_name} filename: 'epoch_{epoch}' monitor: val/loss save_last: True save_top_k: 0 every_n_epochs: 1 progress_bar: _target_: pytorch_lightning.callbacks.TQDMProgressBar refresh_rate: ${computer.progress_bar_refresh_rate} aux_data: [] max_epochs: 100 data_dir: ${root_dir}/datasets root_dir: ${hydra:runtime.cwd} experiment_name: ${dataset.name}__${model.name} mode: train # change that to eval to do the testing num_classes: 0 areas: ['country', 'region', 'sub-region', 'city'] class_name: null streetclip: False blur: False text_tuning: False hydra: run: dir: outputs/${hydra.job.name}/${now:%Y-%m-%d_%H-%M-%S}/${experiment_name} job: chdir: true