Upload EcapaTdnnForSequenceClassification
Browse files- config.json +6 -2
- tdnn_attention.py +1 -1
config.json
CHANGED
|
@@ -1,11 +1,14 @@
|
|
| 1 |
{
|
| 2 |
-
"_attn_implementation_autoset": true,
|
| 3 |
"angular": true,
|
| 4 |
"angular_margin": 0.2,
|
| 5 |
"angular_scale": 30,
|
|
|
|
|
|
|
|
|
|
| 6 |
"attention_channels": 128,
|
| 7 |
"auto_map": {
|
| 8 |
-
"AutoConfig": "configuration_ecapa_tdnn.EcapaTdnnConfig"
|
|
|
|
| 9 |
},
|
| 10 |
"bos_token_id": 1,
|
| 11 |
"decoder_config": {
|
|
@@ -2577,6 +2580,7 @@
|
|
| 2577 |
},
|
| 2578 |
"time_masks": 5,
|
| 2579 |
"time_width": 0.03,
|
|
|
|
| 2580 |
"transformers_version": "4.48.3",
|
| 2581 |
"use_torchaudio": true,
|
| 2582 |
"use_vectorized_spec_augment": true,
|
|
|
|
| 1 |
{
|
|
|
|
| 2 |
"angular": true,
|
| 3 |
"angular_margin": 0.2,
|
| 4 |
"angular_scale": 30,
|
| 5 |
+
"architectures": [
|
| 6 |
+
"EcapaTdnnForSequenceClassification"
|
| 7 |
+
],
|
| 8 |
"attention_channels": 128,
|
| 9 |
"auto_map": {
|
| 10 |
+
"AutoConfig": "configuration_ecapa_tdnn.EcapaTdnnConfig",
|
| 11 |
+
"AutoModelForAudioClassification": "modeling_ecapa_tdnn.EcapaTdnnForSequenceClassification"
|
| 12 |
},
|
| 13 |
"bos_token_id": 1,
|
| 14 |
"decoder_config": {
|
|
|
|
| 2580 |
},
|
| 2581 |
"time_masks": 5,
|
| 2582 |
"time_width": 0.03,
|
| 2583 |
+
"torch_dtype": "float32",
|
| 2584 |
"transformers_version": "4.48.3",
|
| 2585 |
"use_torchaudio": true,
|
| 2586 |
"use_vectorized_spec_augment": true,
|
tdnn_attention.py
CHANGED
|
@@ -276,7 +276,7 @@ class TdnnSeModule(nn.Module):
|
|
| 276 |
def forward(self, inputs, length=None):
|
| 277 |
x = self.group_tdnn_block(inputs)
|
| 278 |
x = self.se_layer(x, length)
|
| 279 |
-
return x +
|
| 280 |
|
| 281 |
|
| 282 |
class Res2NetBlock(nn.Module):
|
|
|
|
| 276 |
def forward(self, inputs, length=None):
|
| 277 |
x = self.group_tdnn_block(inputs)
|
| 278 |
x = self.se_layer(x, length)
|
| 279 |
+
return x + inputs
|
| 280 |
|
| 281 |
|
| 282 |
class Res2NetBlock(nn.Module):
|