From 2ac79cd3f312e485f3fc4f0e63313cc8a3e0bfc6 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期三, 12 六月 2024 19:27:35 +0800
Subject: [PATCH] decoding
---
funasr/bin/train_ds.py | 21 +++++++++++++++++----
1 files changed, 17 insertions(+), 4 deletions(-)
diff --git a/funasr/bin/train_ds.py b/funasr/bin/train_ds.py
index b0931b0..41ecbe4 100644
--- a/funasr/bin/train_ds.py
+++ b/funasr/bin/train_ds.py
@@ -124,14 +124,15 @@
use_ddp=use_ddp,
use_fsdp=use_fsdp,
device=kwargs["device"],
+ excludes=kwargs.get("excludes", None),
output_dir=kwargs.get("output_dir", "./exp"),
**kwargs.get("train_conf"),
)
model = trainer.warp_model(model)
- kwargs["device"] = next(model.parameters()).device
- trainer.device = kwargs["device"]
+ kwargs["device"] = int(os.environ.get("LOCAL_RANK", 0))
+ trainer.device = int(os.environ.get("LOCAL_RANK", 0))
model, optim, scheduler = trainer.warp_optim_scheduler(model, **kwargs)
@@ -158,6 +159,8 @@
time1 = time.perf_counter()
for data_split_i in range(trainer.start_data_split_i, dataloader.data_split_num):
+ time_slice_i = time.perf_counter()
+
dataloader_tr, dataloader_val = dataloader.build_iter(
epoch, data_split_i=data_split_i, start_step=trainer.start_step
)
@@ -178,6 +181,14 @@
torch.cuda.empty_cache()
+ time_escaped = (time.perf_counter() - time_slice_i) / 3600.0
+ logging.info(
+ f"\n\nrank: {local_rank}, "
+ f"time_escaped_epoch: {time_escaped:.3f} hours, "
+ f"estimated to finish {dataloader.data_split_num} data_slices, remaining: {dataloader.data_split_num-data_split_i} slices, {(dataloader.data_split_num-data_split_i)*time_escaped:.3f} hours, "
+ f"epoch: {trainer.max_epoch - epoch} epochs, {((trainer.max_epoch - epoch - 1)*dataloader.data_split_num + dataloader.data_split_num-data_split_i)*time_escaped:.3f} hours\n"
+ )
+
trainer.start_data_split_i = 0
trainer.validate_epoch(model=model, dataloader_val=dataloader_val, epoch=epoch + 1)
scheduler.step()
@@ -189,7 +200,7 @@
time2 = time.perf_counter()
time_escaped = (time2 - time1) / 3600.0
logging.info(
- f"rank: {local_rank}, "
+ f"\n\nrank: {local_rank}, "
f"time_escaped_epoch: {time_escaped:.3f} hours, "
f"estimated to finish {trainer.max_epoch} "
f"epoch: {(trainer.max_epoch - epoch) * time_escaped:.3f} hours\n"
@@ -198,7 +209,9 @@
trainer.train_loss_avg = 0.0
if trainer.rank == 0:
- average_checkpoints(trainer.output_dir, trainer.avg_nbest_model)
+ average_checkpoints(
+ trainer.output_dir, trainer.avg_nbest_model, use_deepspeed=trainer.use_deepspeed
+ )
trainer.close()
--
Gitblit v1.9.1