None type error with local attention

[benderama3]

Trying to run the example from the readme using local attention instead of linear attention. I changed the attention_type and added an additional argument in the TransformerEncoderBuilder.from_kwargs method:

import torch
from fast_transformers.builders import TransformerEncoderBuilder

# Create the builder for our transformers
builder = TransformerEncoderBuilder.from_kwargs(
    n_layers=8,
    n_heads=8,
    query_dimensions=64,
    value_dimensions=64,
    feed_forward_dimensions=1024,
    local_context=8, #ADDED THIS LINE
)

# Build a transformer with softmax attention
builder.attention_type = "full"
softmax_model = builder.get()

# Build a transformer with linear attention
builder.attention_type = "local" #CHANGED THIS LINE
linear_model = builder.get()

# Construct the dummy input
X = torch.rand(10, 1000, 8*64)

# Prepare everythin for CUDA
X = X.cuda()
softmax_model.cuda()
softmax_model.eval()
linear_model.cuda()
linear_model.eval()

# Warmup the GPU
with torch.no_grad():
    softmax_model(X)
    linear_model(X)
torch.cuda.synchronize()

# Measure the execution time
softmax_start = torch.cuda.Event(enable_timing=True)
softmax_end = torch.cuda.Event(enable_timing=True)
linear_start = torch.cuda.Event(enable_timing=True)
linear_end = torch.cuda.Event(enable_timing=True)

with torch.no_grad():
    softmax_start.record()
    y = softmax_model(X)
    softmax_end.record()
    torch.cuda.synchronize()
    print("Softmax: ", softmax_start.elapsed_time(softmax_end), "ms")
    # Softmax: 144 ms (on a GTX1080Ti)

with torch.no_grad():
    linear_start.record()
    y = linear_model(X)
    linear_end.record()
    torch.cuda.synchronize()
    print("Linear: ", linear_start.elapsed_time(linear_end), "ms")
    # Linear: 68 ms (on a GTX1080Ti)

The exemple throws an error:

Traceback (most recent call last):
  File "_.py", line 35, in <module>
    linear_model(X)
  File "/home/jjhon/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/jjhon/.local/lib/python3.6/site-packages/fast_transformers/transformers.py", line 139, in forward
    x = layer(x, attn_mask=attn_mask, length_mask=length_mask)
  File "/home/jjhon/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/jjhon/.local/lib/python3.6/site-packages/fast_transformers/transformers.py", line 81, in forward
    key_lengths=length_mask
  File "/home/jjhon/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/jjhon/.local/lib/python3.6/site-packages/fast_transformers/attention/attention_layer.py", line 109, in forward
    key_lengths
  File "/home/jjhon/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/jjhon/.local/lib/python3.6/site-packages/fast_transformers/attention/local_attention.py", line 82, in forward
    self.local_context
  File "/home/jjhon/.local/lib/python3.6/site-packages/fast_transformers/local_product/__init__.py", line 49, in forward
    local_context
TypeError: 'NoneType' object is not callable

It does work with the other attention modules.
Am I doing something wrong?
Is the local_context argument supposed to be an integer?

Thank you.

EDIT: looks like it is failing using cuda only (pytorch 1.6 with cuda 10.1), it works on the cpu

EDIT2: fixed using --no-cache-dir argument when installing with pip (to recompile)

[15805383399]

I got similar problems
I found that there is no casual_product_gpu in the directory ，and it can't be imported
I have tried --no-cache-dir to recompile, but it didn't work

  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 712, in <module>
    train()
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 587, in train
    losses = net.train_step(batch_x, batch_y, batch_mask)
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 302, in train_step
    h, y_type = self.forward_hidden(x)
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 368, in forward_hidden
    h = self.transformer_encoder(pos_emb, attn_mask)  # y: b x s x d_model
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/transformers.py", line 138, in forward
    x = layer(x, attn_mask=attn_mask, length_mask=length_mask)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/transformers.py", line 77, in forward
    x = x + self.dropout(self.attention(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/attention_layer.py", line 105, in forward
    new_values = self.inner_attention(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/causal_linear_attention.py", line 98, in forward
    V = causal_linear(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/causal_linear_attention.py", line 23, in causal_linear
    V_new = causal_dot_product(Q, K, V)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/causal_product/__init__.py", line 44, in forward
    CausalDotProduct.dot[device.type](
TypeError: 'NoneType' object is not callable

进程已结束，退出代码为 1

[jingshuangliu22]

I got similar problems I found that there is no casual_product_gpu in the directory ，and it can't be imported I have tried --no-cache-dir to recompile, but it didn't work

  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 712, in <module>
    train()
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 587, in train
    losses = net.train_step(batch_x, batch_y, batch_mask)
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 302, in train_step
    h, y_type = self.forward_hidden(x)
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 368, in forward_hidden
    h = self.transformer_encoder(pos_emb, attn_mask)  # y: b x s x d_model
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/transformers.py", line 138, in forward
    x = layer(x, attn_mask=attn_mask, length_mask=length_mask)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/transformers.py", line 77, in forward
    x = x + self.dropout(self.attention(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/attention_layer.py", line 105, in forward
    new_values = self.inner_attention(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/causal_linear_attention.py", line 98, in forward
    V = causal_linear(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/causal_linear_attention.py", line 23, in causal_linear
    V_new = causal_dot_product(Q, K, V)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/causal_product/__init__.py", line 44, in forward
    CausalDotProduct.dot[device.type](
TypeError: 'NoneType' object is not callable

进程已结束，退出代码为 1

Hi, I met the exact same issue and may I ask if you have solved it? @15805383399

[jingshuangliu22]

I got similar problems I found that there is no casual_product_gpu in the directory ，and it can't be imported I have tried --no-cache-dir to recompile, but it didn't work

  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 712, in <module>
    train()
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 587, in train
    losses = net.train_step(batch_x, batch_y, batch_mask)
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 302, in train_step
    h, y_type = self.forward_hidden(x)
  File "/media/lyk/高速/Project/github/compound-word-transformer/workspace/uncond/cp-linear/main-cp.py", line 368, in forward_hidden
    h = self.transformer_encoder(pos_emb, attn_mask)  # y: b x s x d_model
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/transformers.py", line 138, in forward
    x = layer(x, attn_mask=attn_mask, length_mask=length_mask)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/transformers.py", line 77, in forward
    x = x + self.dropout(self.attention(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/attention_layer.py", line 105, in forward
    new_values = self.inner_attention(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/causal_linear_attention.py", line 98, in forward
    V = causal_linear(
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/attention/causal_linear_attention.py", line 23, in causal_linear
    V_new = causal_dot_product(Q, K, V)
  File "/home/lyk/anaconda3/lib/python3.8/site-packages/pytorch_fast_transformers-0.4.0-py3.8-linux-x86_64.egg/fast_transformers/causal_product/__init__.py", line 44, in forward
    CausalDotProduct.dot[device.type](
TypeError: 'NoneType' object is not callable

进程已结束，退出代码为 1

Hi, I met the exact same issue and may I ask if you have solved it? @15805383399

I resolved this one with the methods mentioned in #96 😁