NotImplementedError: Module [ModuleList] is missing the required "forward" function

Question

i get NotImplementedError when try to use Self-Attention on YOLO.

class BertSelfAttention(nn.Module): def init(self, config): super().init() assert config["hidden_size"] % config["num_of_attention_heads"] == 0, "The hidden size is not a multiple of the number of attention heads"

    self.num_attention_heads = config['num_of_attention_heads']
    self.attention_head_size = int(config['hidden_size'] / config['num_of_attention_heads'])
    self.all_head_size = self.num_attention_heads * self.attention_head_size

    self.query = nn.ModuleList([nn.Linear(config['hidden_size'], self.all_head_size)])
    self.key = nn.ModuleList([nn.Linear(config['hidden_size'], self.all_head_size)])
    self.value = nn.ModuleList([nn.Linear(config['hidden_size'], self.all_head_size)])

    self.dense = nn.ModuleList([nn.Linear(config['hidden_size'], config['hidden_size'])])

def transpose_for_scores(self, x):
    new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
    x = x.view(*new_x_shape)
    return x.permute(0, 2, 1, 3)

def forward(self, hidden_states):
    # mixed_query_layer = self.query(hidden_states)  # [Batch_size x Seq_length x Hidden_size]
    # mixed_key_layer = self.key(hidden_states)  # [Batch_size x Seq_length x Hidden_size]
    # mixed_value_layer = self.value(hidden_states)  # [Batch_size x Seq_length x Hidden_size]
    
    size = hidden_states.size(dim=1)
    
    query_layer = self.transpose_for_scores(self.query(torch.randn(16,size)))  # [Batch_size x Num_of_heads x Seq_length x Head_size]
    key_layer = self.transpose_for_scores(self.key(torch.randn(16,size)))  # [Batch_size x Num_of_heads x Seq_length x Head_size]
    value_layer = self.transpose_for_scores(self.value(torch.randn(16,size)))  # [Batch_size x Num_of_heads x Seq_length x Head_size]

    attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,-2))  # [Batch_size x Num_of_heads x Seq_length x Seq_length]
    attention_scores = attention_scores / math.sqrt(self.attention_head_size)  # [Batch_size x Num_of_heads x Seq_length x Seq_length]
    attention_probs = nn.Softmax(dim=-1)(attention_scores)  # [Batch_size x Num_of_heads x Seq_length x Seq_length]
    context_layer = torch.matmul(attention_probs, value_layer)  # [Batch_size x Num_of_heads x Seq_length x Head_size]

    context_layer = context_layer.permute(0, 2, 1, 3).contiguous()  # [Batch_size x Seq_length x Num_of_heads x Head_size]
    new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)  # [Batch_size x Seq_length x Hidden_size]
    context_layer = context_layer.view(*new_context_layer_shape)  # [Batch_size x Seq_length x Hidden_size]

    output = self.dense(context_layer)

    return output

That's the result :(

Training start...

 Epoch        lr  iou_loss  dfl_loss  cls_loss

0%| | 0/38 [00:11<?, ?it/s]
ERROR in training steps. ERROR in training loop or eval/save model. Traceback (most recent call last): File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/tools/train.py", line 143, in main(args) File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/tools/train.py", line 133, in main trainer.train() File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/yolov6/core/engine.py", line 121, in train self.train_one_epoch(self.epoch) File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/yolov6/core/engine.py", line 135, in train_one_epoch self.train_in_steps(epoch_num, self.step) File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/yolov6/core/engine.py", line 152, in train_in_steps preds, s_featmaps = self.model(images) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in _call_impl return forward_call(*args, **kwargs) File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/yolov6/models/yolo.py", line 36, in forward x = self.neck(x) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in _call_impl return forward_call(*args, **kwargs) File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/yolov6/models/reppan.py", line 362, in forward fpn_out0 = self.selfattention(x0) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in _call_impl return forward_call(*args, **kwargs) File "/content/drive/MyDrive/KULIAH/SKRIPSI/YOLOv6/yolov6/layers/common.py", line 50, in forward mixed_query_layer = self.query(hidden_states) # [Batch_size x Seq_length x Hidden_size] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in _call_impl return forward_call(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 372, in _forward_unimplemented raise NotImplementedError(f"Module [{type(self).name}] is missing the required "forward" function") NotImplementedError: Module [ModuleList] is missing the required "forward" function

Answer 1

A ModuleList is just a list that tracks pytorch objects/parameters. You can't call it because it has no forward method. I'm not sure why you are putting a single pytorch module inside a ModuleList - you can just have the module on its own.

If you are looking for a pure pytorch implementation of multihead attention, you can check this implementation