submitit/submitit_train.py [78:110]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        }
    grid = {k:v[0] for k, v in grid_meta.items()}
    save_key = {k:v[1] for k, v in grid_meta.items()}

    hyper_parameters = list(grid_parameters(grid))
    jobs = []
    for i, grid_data in enumerate(hyper_parameters):
        cluster_cfg = cluster_cfg._replace(dist_url=get_init_file().as_uri())
        train_cfg = train_cfg._replace(**grid_data)

        run_name = f"{train_cfg.prefix}"
        for k, v in grid_data.items():
            run_name += "-" + save_key[k](v)
        train_cfg = train_cfg._replace(output_dir=os.path.join(log_dir, run_name))

        # Chronos needs a different job name each time
        executor.update_parameters(name=f"sweep_{i:02d}_{uuid.uuid4().hex}")
        trainer = Trainer(train_cfg, cluster_cfg)
        job = executor.submit(trainer)
        jobs.append(job)
        print(f"Run {i:02d} submitted with train cfg: {train_cfg}, cluster cfg: {cluster_cfg}")
    print(f"Submitted jobs ids: {','.join([str(job.job_id) for job in jobs])}")

    # Wait for the master's results of each job
    results = [job.task(0).result() for job in jobs]
    print(f"Jobs results: {results}")
    best_job = np.argmax(results)
    print(f"Best configuration: {hyper_parameters[best_job]} (val acc = {results[best_job]:.1%})")


if __name__ == "__main__":
    args = train_args()
    grid_search(args)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



submitit/submitit_train_qa.py [88:120]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        }
    grid = {k:v[0] for k, v in grid_meta.items()}
    save_key = {k:v[1] for k, v in grid_meta.items()}
    
    hyper_parameters = list(grid_parameters(grid))
    jobs = []
    for i, grid_data in enumerate(hyper_parameters):
        cluster_cfg = cluster_cfg._replace(dist_url=get_init_file().as_uri())
        train_cfg = train_cfg._replace(**grid_data)

        run_name = f"{train_cfg.prefix}"
        for k, v in grid_data.items():
            run_name += "-" + save_key[k](v)
        train_cfg = train_cfg._replace(output_dir=os.path.join(log_dir, run_name))

        # Chronos needs a different job name each time
        executor.update_parameters(name=f"sweep_{i:02d}_{uuid.uuid4().hex}")
        trainer = Trainer(train_cfg, cluster_cfg)
        job = executor.submit(trainer)
        jobs.append(job)
        print(f"Run {i:02d} submitted with train cfg: {train_cfg}, cluster cfg: {cluster_cfg}")
    print(f"Submitted jobs ids: {','.join([str(job.job_id) for job in jobs])}")

    # Wait for the master's results of each job
    results = [job.task(0).result() for job in jobs]
    print(f"Jobs results: {results}")
    best_job = np.argmax(results)
    print(f"Best configuration: {hyper_parameters[best_job]} (val acc = {results[best_job]:.1%})")


if __name__ == "__main__":
    args = train_args()
    grid_search(args)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -