test wav2vec KOR Zeroth finetuned HuggingFace

config.json

@@ -0,0 +1,107 @@
+{
+  "_name_or_path": "/nfs/datab/longnv/wav2vec2-xls-r-300m-korean.bin",
+  "activation_dropout": 0.1,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "classifier_proj_size": 256,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.0,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.0,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.0,
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.25,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.75,
+  "model_type": "wav2vec2",
+  "num_adapter_layers": 3,
+  "num_attention_heads": 16,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "num_negatives": 100,
+  "output_hidden_size": 1024,
+  "pad_token_id": 1204,
+  "proj_codevector_dim": 768,
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.17.0.dev0",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 1207,
+  "xvector_output_dim": 512
+}

models/sslassist.py

@@ -7,7 +7,7 @@
 import torch.nn.functional as F
 from torch import Tensor
 import fairseq
-
+from transformers import AutoModel
 
 ___author__ = "Hemlata Tak"
 __email__ = "[email protected]"
@@ -21,9 +21,11 @@ class SSLModel(nn.Module):
     def __init__(self,device):
         super(SSLModel, self).__init__()
 
-        cp_path = '/datac/longnv/SSL_Anti-spoofing/pretrained/xlsr2_300m.pt'   # Change the pre-trained XLSR model path. 
-        model, cfg, task = fairseq.checkpoint_utils.load_model_ensemble_and_task([cp_path])
-        self.model = model[0]
+        cp_path = '/nfs/datab/longnv/wav2vec2-xls-r-300m-korean.bin'   # Change the pre-trained XLSR model path. 
+        # model, cfg, task = fairseq.checkpoint_utils.load_model_ensemble_and_task([cp_path])
+        model = AutoModel.from_pretrained("./")
+        # self.model = model[0]
+        self.model = model
         self.device=device
         self.out_dim = 1024
         return
@@ -45,7 +47,8 @@ def extract_feat(self, input_data):
                 input_tmp = input_data
 
             # [batch, length, dim]
-            emb = self.model(input_tmp, mask=False, features_only=True)['x']
+            # emb = self.model(input_tmp, mask=False, features_only=True)['x']
+            emb = self.model(input_tmp).last_hidden_state
         return emb
 
 

oc_training.py

@@ -336,7 +336,7 @@ def collate_fn(self, batch):
 
 
     # WandB – Initialize a new run
-    wandb.init(project="oc_classifier", entity="longnv")
+    wandb.init(project="oc_classifier-kor", entity="longnv")
 
     # Number of epochs
     num_epochs = 100