From 6ce4909ecc39b990e253d692afc4b28451cde33e Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期五, 22 三月 2024 21:08:55 +0800
Subject: [PATCH] update

---
 funasr/train_utils/trainer.py |   30 ++++++++++++++++--------------
 1 files changed, 16 insertions(+), 14 deletions(-)

diff --git a/funasr/train_utils/trainer.py b/funasr/train_utils/trainer.py
index c443c6f..3e83581 100644
--- a/funasr/train_utils/trainer.py
+++ b/funasr/train_utils/trainer.py
@@ -198,6 +198,8 @@
                 for k in dst_state.keys():
                     if not k.startswith("module.") and "module."+k in src_state.keys():
                         k_ddp = "module."+k
+                    elif k.startswith("module.") and "module."+k not in src_state.keys():
+                        k_ddp = k.replace("module.", "", 1)
                     else:
                         k_ddp = k
                     if k_ddp in src_state.keys():
@@ -288,13 +290,13 @@
                 self.train_loss_avg = (self.train_loss_avg*batch_idx + loss.detach().cpu().item())/(batch_idx+1)
                 if "acc" in stats:
                     self.train_acc_avg = (self.train_acc_avg * batch_idx + stats["acc"].detach().cpu().item()) / (batch_idx + 1)
-                if self.use_ddp or self.use_fsdp:
-                    train_loss_avg = torch.tensor(self.train_loss_avg, dtype=torch.float32).to(self.device)
-                    train_acc_avg = torch.tensor(self.train_acc_avg, dtype=torch.float32).to(self.device)
-                    dist.all_reduce(train_loss_avg, op=dist.ReduceOp.SUM)
-                    dist.all_reduce(train_acc_avg, op=dist.ReduceOp.SUM)
-                    self.train_loss_avg = train_loss_avg.detach().cpu().item() / self.world_size
-                    self.train_acc_avg = train_acc_avg.detach().cpu().item() / self.world_size
+                # if self.use_ddp or self.use_fsdp:
+                #     train_loss_avg = torch.tensor(self.train_loss_avg, dtype=torch.float32).to(self.device)
+                #     train_acc_avg = torch.tensor(self.train_acc_avg, dtype=torch.float32).to(self.device)
+                #     dist.all_reduce(train_loss_avg, op=dist.ReduceOp.SUM)
+                #     dist.all_reduce(train_acc_avg, op=dist.ReduceOp.SUM)
+                #     self.train_loss_avg = train_loss_avg.detach().cpu().item() / self.world_size
+                #     self.train_acc_avg = train_acc_avg.detach().cpu().item() / self.world_size
                 
             
             # Perform an optimizer step only after accumulating enough gradients
@@ -410,13 +412,13 @@
                 self.val_loss_avg = (self.val_loss_avg*batch_idx + loss.detach().cpu().item())/(batch_idx+1)
                 if "acc" in stats:
                     self.val_acc_avg = (self.val_acc_avg * batch_idx + stats["acc"].detach().cpu().item()) / (batch_idx + 1)
-                if self.use_ddp or self.use_fsdp:
-                    val_loss_avg = torch.tensor(self.val_loss_avg, dtype=torch.float32).to(self.device)
-                    val_acc_avg = torch.tensor(self.val_acc_avg, dtype=torch.float32).to(self.device)
-                    dist.all_reduce(val_loss_avg, op=dist.ReduceOp.SUM)
-                    dist.all_reduce(val_acc_avg, op=dist.ReduceOp.SUM)
-                    self.val_loss_avg = val_loss_avg.detach().cpu().item() / self.world_size
-                    self.val_acc_avg = val_acc_avg.detach().cpu().item() / self.world_size
+                # if self.use_ddp or self.use_fsdp:
+                #     val_loss_avg = torch.tensor(self.val_loss_avg, dtype=torch.float32).to(self.device)
+                #     val_acc_avg = torch.tensor(self.val_acc_avg, dtype=torch.float32).to(self.device)
+                #     dist.all_reduce(val_loss_avg, op=dist.ReduceOp.SUM)
+                #     dist.all_reduce(val_acc_avg, op=dist.ReduceOp.SUM)
+                #     self.val_loss_avg = val_loss_avg.detach().cpu().item() / self.world_size
+                #     self.val_acc_avg = val_acc_avg.detach().cpu().item() / self.world_size
                 
                 batch_num_epoch = -1
                 if hasattr(dataloader_val, "__len__"):

--
Gitblit v1.9.1