From 9b4e9cc8a0311e5243d69b73ed073e7ea441982e Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期三, 27 三月 2024 16:05:29 +0800
Subject: [PATCH] train update
---
funasr/bin/train.py | 23 ++++++++++++-----------
1 files changed, 12 insertions(+), 11 deletions(-)
diff --git a/funasr/bin/train.py b/funasr/bin/train.py
index 5cf54da..880bb63 100644
--- a/funasr/bin/train.py
+++ b/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,
@@ -176,12 +176,12 @@
# 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)
+ dataloader_tr, dataloader_val = dataloader.build_iter(epoch)
trainer.train_epoch(
model=model,
optim=optim,
@@ -192,13 +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)
@@ -213,7 +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()
--
Gitblit v1.9.1