update train recipe

游雁

2024-02-21 26ab38b56cb4f69cc82e5d58907f2a57f6f2cbdd

update train recipe

 examples/aishell/branchformer/run.sh

@@ -174,6 +174,7 @@
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++ncpu=1 \
          ++disable_log=true \
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&


 examples/aishell/conformer/run.sh

@@ -173,6 +173,7 @@
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++ncpu=1 \
          ++disable_log=true \
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&


 examples/aishell/e_branchformer/run.sh

@@ -174,6 +174,7 @@
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++ncpu=1 \
          ++disable_log=true \
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&


 examples/aishell/paraformer/run.sh

@@ -173,6 +173,7 @@
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++ncpu=1 \
          ++disable_log=true \
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&


 examples/aishell/transformer/run.sh

@@ -174,6 +174,7 @@
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++ncpu=1 \
          ++disable_log=true \
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&


 funasr/train_utils/trainer.py

@@ -3,6 +3,7 @@
import torch
import logging
from tqdm import tqdm
from datetime import datetime
import torch.distributed as dist
from contextlib import nullcontext
# from torch.utils.tensorboard import SummaryWriter
@@ -283,7 +284,10 @@
                                             torch.cuda.max_memory_reserved()/1024/1024/1024,
                                             )
                lr = self.scheduler.get_last_lr()[0]
                time_now = datetime.now()
                time_now = now.strftime("%Y-%m-%d %H:%M:%S")
                description = (
                    f"{time_now}, "
                    f"rank: {self.local_rank}, "
                    f"epoch: {epoch}/{self.max_epoch}, "
                    f"step: {batch_idx+1}/{len(self.dataloader_train)}, total: {self.batch_total}, "