Upload EcapaTdnnForSequenceClassification

config.json

@@ -1,11 +1,14 @@
 {
-  "_attn_implementation_autoset": true,
   "angular": true,
   "angular_margin": 0.2,
   "angular_scale": 30,
+  "architectures": [
+    "EcapaTdnnForSequenceClassification"
+  ],
   "attention_channels": 128,
   "auto_map": {
-    "AutoConfig": "configuration_ecapa_tdnn.EcapaTdnnConfig"
+    "AutoConfig": "configuration_ecapa_tdnn.EcapaTdnnConfig",
+    "AutoModelForAudioClassification": "modeling_ecapa_tdnn.EcapaTdnnForSequenceClassification"
   },
   "bos_token_id": 1,
   "decoder_config": {
@@ -2577,6 +2580,7 @@
   },
   "time_masks": 5,
   "time_width": 0.03,
+  "torch_dtype": "float32",
   "transformers_version": "4.48.3",
   "use_torchaudio": true,
   "use_vectorized_spec_augment": true,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e437b7e91fe5f7a7a3012f7afb06cb155c7a01cc0662e249cffb4f7a6cd6b52
+oid sha256:4da89b0b6d405974f1e332bdc9945fae76222d7ddf0f955653fba9a00cca0339
 size 26039912

tdnn_attention.py

@@ -273,8 +273,8 @@ class TdnnSeModule(nn.Module):
 
         self.apply(lambda x: init_weights(x, mode=init_mode))
 
-    def forward(self, input, length=None):
-        x = self.group_tdnn_block(input)
+    def forward(self, inputs, length=None):
+        x = self.group_tdnn_block(inputs)
         x = self.se_layer(x, length)
         return x + input