train update

游雁

2024-03-27 9b4e9cc8a0311e5243d69b73ed073e7ea441982e

 funasr/bin/train.py

@@ -128,7 +128,8 @@
    else:
        model = model.to(device=kwargs.get("device", "cuda"))

    logging.info(f"{model}")
    if local_rank == 0:
        logging.info(f"{model}")
    kwargs["device"] = next(model.parameters()).device
        
    # optim
@@ -149,8 +150,8 @@
    # dataset
    logging.info("Build dataloader")
    dataloader_class = tables.dataloader_classes.get(kwargs["dataset_conf"].get("dataloader", "DataloaderMapStyle"))
    dataloader_tr, dataloader_val = dataloader_class(**kwargs)

    dataloader = dataloader_class(**kwargs)
    # dataloader_tr, dataloader_val = dataloader_class(**kwargs)
    trainer = Trainer(local_rank=local_rank,
                      use_ddp=use_ddp,
                      use_fsdp=use_fsdp,
@@ -172,15 +173,15 @@
    except:
        writer = None

    if use_ddp or use_fsdp:
        context = Join([model])
    else:
        context = nullcontext()

    # if use_ddp or use_fsdp:
    #     context = Join([model])
    # else:
    #     context = nullcontext()
    context = nullcontext()
    for epoch in range(trainer.start_epoch, trainer.max_epoch + 1):
        time1 = time.perf_counter()
        with context:
            
            dataloader_tr, dataloader_val = dataloader.build_iter(epoch)
            trainer.train_epoch(
                                model=model,
                                optim=optim,
@@ -191,13 +192,14 @@
                                epoch=epoch,
                                writer=writer
                                )
        with context:
            trainer.validate_epoch(
                model=model,
                dataloader_val=dataloader_val,
                epoch=epoch,
                writer=writer
            )
        scheduler.step()
        trainer.validate_epoch(
            model=model,
            dataloader_val=dataloader_val,
            epoch=epoch,
            writer=writer
        )

        
        trainer.save_checkpoint(epoch, model=model, optim=optim, scheduler=scheduler, scaler=scaler)
@@ -212,7 +214,7 @@


    if trainer.rank == 0:
        average_checkpoints(trainer.output_dir, trainer.avg_nbest_model, trainer.val_acc_list)
        average_checkpoints(trainer.output_dir, trainer.avg_nbest_model)

    trainer.close()