Error occurred when changing the feature extraction method to wavlm. #39
Description
Thank you so much for sharing your research findings.
The feature extraction method based on Fbank was trained normally and I was able to confirm results similar to the results you shared.
<cam++.yml>
- use_hf_model: True
- feature_method: 'microsoft/wavlm-base-plus'
However, when changing the feature extraction method to wavlm, there are cases where training is suddenly stopped.
I would like to ask for advice on the error.
`2025-08-11 02:04:38.377 | INFO | macls.optimizer:build_optimizer:16 - 成功创建优化方法：Adam，参数为：{'lr': 0.001, 'weight_decay': 1e-05}
2025-08-11 02:04:38.377 | INFO | macls.optimizer:build_lr_scheduler:31 - 成功创建学习率衰减：WarmupCosineSchedulerLR，参数为：{'min_lr': 1e-05, 'max_lr': 0.001, 'warmup_epoch': 5, 'fix_epoch': 60, 'step_per_epoch': 245}
2025-08-11 02:04:38.377 | INFO | macls.trainer:train:335 - 训练数据：7858
/root/miniconda3/lib/python3.12/site-packages/torch/nn/functional.py:5849: UserWarning: Support for mismatched key_padding_mask and attn_mask is deprecated. Use same type for both instead.
warnings.warn(
2025-08-11 02:04:42.653 | INFO | macls.trainer:__train_epoch:278 - Train epoch: [1/60], batch: [1/245], loss: 2.78939, accuracy: 0.09375, learning rate: 0.00001000, speed: 1.87 data/sec, eta: 17:27:25
2025-08-11 02:05:16.676 | INFO | macls.trainer:__train_epoch:278 - Train epoch: [1/60], batch: [11/245], loss: 3.13291, accuracy: 0.26562, learning rate: 0.00001808, speed: 2.35 data/sec, eta: 13:52:54
2025-08-11 02:05:50.317 | INFO | macls.trainer:__train_epoch:278 - Train epoch: [1/60], batch: [21/245], loss: 2.62199, accuracy: 0.30000, learning rate: 0.00002616, speed: 2.38 data/sec, eta: 13:43:00
2025-08-11 02:06:26.822 | INFO | macls.trainer:__train_epoch:278 - Train epoch: [1/60], batch: [31/245], loss: 2.13521, accuracy: 0.32500, learning rate: 0.00003424, speed: 2.19 data/sec, eta: 14:52:28
2025-08-11 02:07:00.074 | INFO | macls.trainer:__train_epoch:278 - Train epoch: [1/60], batch: [41/245], loss: 1.89144, accuracy: 0.39062, learning rate: 0.00004233, speed: 2.41 data/sec, eta: 13:32:23
Traceback (most recent call last):
File "/mnt/nas5/docker/AudioClassification-Pytorch/train.py", line 27, in
trainer.train(save_model_path=args.save_model_path,
File "/mnt/nas5/docker/AudioClassification-Pytorch/macls/trainer.py", line 353, in train
self.__train_epoch(epoch_id=epoch_id, local_rank=local_rank, writer=writer, nranks=nranks)
File "/mnt/nas5/docker/AudioClassification-Pytorch/macls/trainer.py", line 231, in __train_epoch
for batch_id, (features, label, input_len) in enumerate(self.train_loader):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/site-packages/torch/utils/data/dataloader.py", line 701, in next
data = self._next_data()
^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/site-packages/torch/utils/data/dataloader.py", line 757, in _next_data
data = self._dataset_fetcher.fetch(index) # may raise StopIteration
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/site-packages/torch/utils/data/_utils/fetch.py", line 52, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
~~~~~~~~~~~~^^^^^
File "/mnt/nas5/docker/AudioClassification-Pytorch/macls/data_utils/reader.py", line 100, in getitem
feature = self.spec_augment(feature.cpu().numpy())
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/site-packages/yeaudio/augmentation.py", line 291, in call
x = self.time_mask(x)
^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/site-packages/yeaudio/augmentation.py", line 273, in time_mask
length = random.randint(1, mask_time_len)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/random.py", line 336, in randint
return self.randrange(a, b+1)
^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.12/random.py", line 319, in randrange
raise ValueError(f"empty range in randrange({start}, {stop})")
ValueError: empty range in randrange(1, 1)`