fix style

egs/audioset/AT/zipformer/at_datamodule.py

@@ -17,7 +17,6 @@
 import argparse
 import inspect
 import logging
-import pickle
 from functools import lru_cache
 from pathlib import Path
 from typing import Any, Dict, Optional

egs/audioset/AT/zipformer/evaluate.py

@@ -160,8 +160,6 @@ def decode_dataset(
     model: nn.Module,
 ) -> Dict:
     num_cuts = 0
-    embedding_dict = {}
-    teacher_embedding_dict = {}
 
     try:
         num_batches = len(dl)

egs/audioset/AT/zipformer/export-onnx.py

@@ -62,7 +62,7 @@ use the exported ONNX models.
 import argparse
 import logging
 from pathlib import Path
-from typing import Dict, Tuple
+from typing import Dict
 
 import k2
 import onnx
@@ -189,9 +189,9 @@ class OnnxAudioTagger(nn.Module):
           x_lens:
             A 1-D tensor of shape (N,). Its dtype is torch.int64
         Returns:
-          Return a tuple containing:
-            - encoder_out, A 3-D tensor of shape (N, T', joiner_dim)
-            - encoder_out_lens, A 1-D tensor of shape (N,)
+          Return a tensor containing:
+            - logits, A 2-D tensor of shape (N, num_classes)
+
         """
         x, x_lens = self.encoder_embed(x, x_lens)
         src_key_padding_mask = make_pad_mask(x_lens)
@@ -200,13 +200,12 @@ class OnnxAudioTagger(nn.Module):
         encoder_out = encoder_out.permute(1, 0, 2)  # (N,T,C)
 
         logits = self.classifier(encoder_out)  # (N, T, num_classes)
-        padding_mask = make_pad_mask(encoder_out_lens)
-        logits[padding_mask] = 0
-        logits = logits.sum(dim=1)  # mask the padding frames
-        logits = logits / (~padding_mask).sum(dim=1).unsqueeze(-1).expand_as(
-            logits
-        )  # normalize the logits
-        print(logits.shape)
+        # Note that this is slightly different from model.py for better
+        # support of onnx
+        N = logits.shape[0]
+        for i in range(N):
+            logits[i, encoder_out_lens[i] :] = 0
+        logits = logits.sum(dim=1) / encoder_out_lens.unsqueeze(-1)
         return logits
 
 
@@ -237,7 +236,7 @@ def export_audio_tagging_model_onnx(
     x = torch.zeros(1, 200, 80, dtype=torch.float32)
     x_lens = torch.tensor([200], dtype=torch.int64)
 
-    model = torch.jit.script(model)
+    model = torch.jit.trace(model, (x, x_lens))
 
     torch.onnx.export(
         model,

egs/audioset/AT/zipformer/onnx_pretrained.py

@@ -135,7 +135,6 @@ class OnnxModel:
         meta = self.model.get_modelmeta().custom_metadata_map
         print(meta)
 
-
     def __call__(
         self,
         x: torch.Tensor,
@@ -162,6 +161,7 @@ class OnnxModel:
         )
         return torch.from_numpy(out[0])
 
+
 def read_sound_files(
     filenames: List[str], expected_sample_rate: float
 ) -> List[torch.Tensor]:
@@ -232,7 +232,7 @@ def main():
 
     feature_lengths = torch.tensor(feature_lengths, dtype=torch.int64)
     logits = model(features, feature_lengths)
-    
+
     for filename, logit in zip(args.sound_files, logits):
         topk_prob, topk_index = logit.sigmoid().topk(5)
         topk_labels = [label_dict[index.item()] for index in topk_index]