finetune

游雁

2024-03-24 ed952ff630a1ad89df488d3c657ef736129e6c10

finetune

 funasr/bin/train.py

@@ -150,8 +150,8 @@
    # dataset
    logging.info("Build dataloader")
    dataloader_class = tables.dataloader_classes.get(kwargs["dataset_conf"].get("dataloader", "DataloaderMapStyle"))
    # dataloader = dataloader_class(**kwargs)
    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,
@@ -181,7 +181,7 @@
    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)
            dataloader_tr, dataloader_val = dataloader.build_iter(epoch)
            trainer.train_epoch(
                                model=model,
                                optim=optim,

 funasr/datasets/dataloader_entry.py

@@ -4,7 +4,7 @@

from funasr.register import tables

@tables.register("dataloader_classes", "DataloaderMapStyle")
# @tables.register("dataloader_classes", "DataloaderMapStyle")
def DataloaderMapStyle(frontend=None, tokenizer=None, **kwargs):
    # dataset
    logging.info("Build dataloader")
@@ -25,7 +25,7 @@
    
    return dataloader_tr, dataloader_val

# @tables.register("dataloader_classes", "DataloaderMapStyle")
@tables.register("dataloader_classes", "DataloaderMapStyle")
class DataloaderMapStyle:
    def __init__(self, frontend=None, tokenizer=None, **kwargs):
        # dataset

 funasr/train_utils/trainer.py

@@ -371,8 +371,7 @@
                
        if self.use_ddp or self.use_fsdp:
            dist.barrier()

        iterator_stop = torch.tensor(0).to(self.device)
            iterator_stop = torch.tensor(0).to(self.device)
        
        

@@ -402,12 +401,10 @@
            iterator_stop = torch.tensor(0).to(self.device)
            dataloader_val.batch_sampler.set_epoch(epoch)
            for batch_idx, batch in enumerate(dataloader_val):
                # if self.use_ddp or self.use_fsdp:
                #     dist.all_reduce(iterator_stop, dist.ReduceOp.SUM)
                #     if epoch >= 1:
                #         print(f"iterator_stop: {iterator_stop}\n")
                #     if iterator_stop > 0:
                #         break
                if self.use_ddp or self.use_fsdp:
                    dist.all_reduce(iterator_stop, dist.ReduceOp.SUM)
                    if iterator_stop > 0:
                        break
                time1 = time.perf_counter()
                speed_stats["data_load"] = f"{time1 - time5:0.3f}"
                batch = to_device(batch, self.device)
@@ -467,7 +464,7 @@

        if self.use_ddp or self.use_fsdp:
            dist.barrier()
        iterator_stop = torch.tensor(0).to(self.device)
            iterator_stop = torch.tensor(0).to(self.device)
        
        
    def log(self,