dpr_scale/generate_embeddings.py [14:30]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    task = hydra.utils.instantiate(cfg.task, _recursive_=False)
    transform = hydra.utils.instantiate(cfg.task.transform)
    datamodule = hydra.utils.instantiate(cfg.datamodule, transform=transform)

    # trainer.fit does some setup, so we need to call it even though no training is done
    with open_dict(cfg):
        cfg.trainer.limit_train_batches = 0
        if "plugins" in cfg.trainer:
            cfg.trainer.pop("plugins")  # remove ddp_sharded, since it breaks during loading
    print(cfg)
    trainer = Trainer(**cfg.trainer)
    trainer.fit(task, datamodule=datamodule)
    trainer.test(task, datamodule=datamodule)


if __name__ == "__main__":
    main()
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



dpr_scale/generate_query_embeddings.py [16:32]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    task = hydra.utils.instantiate(cfg.task, _recursive_=False)
    transform = hydra.utils.instantiate(cfg.task.transform)
    datamodule = hydra.utils.instantiate(cfg.datamodule, transform=transform)

    # trainer.fit does some setup, so we need to call it even though no training is done
    with open_dict(cfg):
        cfg.trainer.limit_train_batches = 0
        if "plugins" in cfg.trainer:
            cfg.trainer.pop("plugins")  # remove ddp_sharded, since it breaks during loading
    print(cfg)
    trainer = Trainer(**cfg.trainer)
    trainer.fit(task, datamodule=datamodule)
    trainer.test(task, datamodule=datamodule)


if __name__ == "__main__":
    main()
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -