Views
No views yet
| Base | Large | ||
| CNN Encoder | strides | 5, 2, 2, 2, 2, 2, 2 | |
| kernel width | 10, 3, 3, 3, 3, 2, 2 | ||
| channel | 512 | ||
| Transformer Encoder | Layer | 12 | 24 |
| embedding dim | 768 | 1024 | |
| inner FFN dim | 3072 | 4096 | |
| attention heads | 8 | 16 | |
| Projection | dim | 256 | 768 |
| Params | 95M | 317M |
1import torch
2from transformers import HubertModel
3
4model = HubertModel.from_pretrained("team-lucid/hubert-large-korean")
5
6wav = torch.ones(1, 16000)
7outputs = model(wav)
8print(f"Input: {wav.shape}") # [1, 16000]
9print(f"Output: {outputs.last_hidden_state.shape}") # [1, 49, 768]1import jax.numpy as jnp
2from transformers import FlaxAutoModel
3
4model = FlaxAutoModel.from_pretrained("team-lucid/hubert-large-korean", trust_remote_code=True)
5
6wav = jnp.ones((1, 16000))
7outputs = model(wav)
8print(f"Input: {wav.shape}") # [1, 16000]
9print(f"Output: {outputs.last_hidden_state.shape}") # [1, 49, 768]| Hyperparameter | Base | Large |
|---|---|---|
| Warmup Steps | 32,000 | 32,000 |
| Learning Rates | 5e-4 | 1.5e-3 |
| Batch Size | 128 | 128 |
| Weight Decay | 0.01 | 0.01 |
| Max Steps | 400,000 | 400,000 |
| Learning Rate Decay | 0.1 | 0.1 |
| \(Adam\beta_1\) | 0.9 | 0.9 |
| \(Adam\beta_2\) | 0.99 | 0.99 |