update

游雁

2024-03-22 fffb628d31f8d019bd9af846400e4cb6e6e874fa

update

 funasr/train_utils/trainer.py

@@ -248,8 +248,12 @@
        optim.zero_grad()
        speed_stats = {}
        time5 = time.perf_counter()
        
        iterator_stop = torch.tensor(0).to(self.device)
        for batch_idx, batch in enumerate(dataloader_train):
            if self.use_ddp or self.use_fsdp:
                dist.all_reduce(iterator_stop, dist.ReduceOp.SUM)
                if iterator_stop > 0:
                    break
            self.batch_total += 1
            time1 = time.perf_counter()
            speed_stats["data_load"] = f"{time1-time5:0.3f}"
@@ -356,7 +360,11 @@
            if (batch_idx+1) % self.save_checkpoint_interval == 0:
                self.save_checkpoint(epoch, model=model, optim=optim, scheduler=scheduler, scaler=scaler, step=batch_idx+1)

        
        else:
            if self.use_ddp or self.use_fsdp:
                iterator_stop.fill_(1)
                dist.all_reduce(iterator_stop, dist.ReduceOp.SUM)
                
        if self.use_ddp or self.use_fsdp:
            dist.barrier()
        
@@ -383,7 +391,12 @@
            
            speed_stats = {}
            time5 = time.perf_counter()
            iterator_stop = torch.tensor(0).to(self.device)
            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 iterator_stop > 0:
                        break
                time1 = time.perf_counter()
                speed_stats["data_load"] = f"{time1 - time5:0.3f}"
                batch = to_device(batch, self.device)
@@ -433,9 +446,16 @@
                         tag="val",
                         )

            else:
                if self.use_ddp or self.use_fsdp:
                    iterator_stop.fill_(1)
                    dist.all_reduce(iterator_stop, dist.ReduceOp.SUM)
                    
        self.val_acc_list.append(self.val_acc_avg)
        model.train()
        



        if self.use_ddp or self.use_fsdp:
            dist.barrier()