Hugging Face's logo

lmzjms
/
wavlm-large

PyTorch
JAX
Model card Files Community
wavlm-large
File size: 817 Bytes 
0b32ad6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
 

# this voxceleb1 is doing speaker classification task!
runner:
  total_steps: 200000
  gradient_clipping: 1
  gradient_accumulate_steps: 4

  log_step: 500
  eval_step: 1000
  save_step: 1000
  max_keep: 4
  eval_dataloaders: 
    - dev
    - test
  
optimizer:
  name: TorchOptim
  torch_optim_name: Adam
  lr: 1.0e-4

# # comment the whole scheduler config block to disable learning rate scheduling
# scheduler:
#   name: linear_schedule_with_warmup
#   num_warmup_steps: 5000

downstream_expert: 
  datarc:
    file_path: /groups/public/benchmark_datasets/speaker_classifi/wav
    meta_data: /groups/public/benchmark_datasets/speaker_classifi/veri_test_class.txt
    num_workers: 12
    train_batch_size: 8
    eval_batch_size: 1
    max_timestep: 128000

  modelrc:
    projector_dim: 256
    select: FrameLevel