Models

Networks

---

source

FC

 FC (input_size:int, output_size:int, hidden_sizes:List[int],
     activation:torch.nn.modules.module.Module=LeakyReLU(negative_slope=0.
     2, inplace=True), layer_norm:bool=False)

A fully connected network with an activation function and optional layer normalization

	Type	Default	Details
input_size	int		The size of the input
output_size	int		The size of the output
hidden_sizes	typing.List[int]		The sizes of the hidden layers
activation	Module	LeakyReLU(negative_slope=0.2, inplace=True)
layer_norm	bool	False	Whether to use layer normalization

---

source

FCBlock

 FCBlock (input_size:int, output_size:int,
          activation:torch.nn.modules.module.Module=LeakyReLU(negative_slo
          pe=0.2, inplace=True), layer_norm:bool=False)

A fully connected block with an activation function and optional layer normalization

	Type	Default	Details
input_size	int		The size of the input
output_size	int		The size of the output
activation	Module	LeakyReLU(negative_slope=0.2, inplace=True)
layer_norm	bool	False	Whether to use layer normalization

Test

model = FC(
    input_size=7,
    output_size=100,
    hidden_sizes=[100, 100],
)

x = torch.randn(1, 7)
y = model(x)

assert y.shape == torch.Size([1, 100])
summary(model, input_size=(1, 7))

/home/diaz/anaconda3/envs/modal/lib/python3.10/site-packages/torchinfo/torchinfo.py:477: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  action_fn=lambda data: sys.getsizeof(data.storage()),
/home/diaz/anaconda3/envs/modal/lib/python3.10/site-packages/torch/storage.py:665: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  return super().__sizeof__() + self.nbytes()

==========================================================================================
Layer (type:depth-idx)                   Output Shape              Param #
==========================================================================================
FC                                       [1, 100]                  --
├─Sequential: 1-1                        [1, 100]                  --
│    └─FCBlock: 2-1                      [1, 100]                  --
│    │    └─Linear: 3-1                  [1, 100]                  800
│    │    └─Identity: 3-2                [1, 100]                  --
│    └─FCBlock: 2-4                      --                        (recursive)
│    │    └─LeakyReLU: 3-3               [1, 100]                  --
│    └─FCBlock: 2-3                      [1, 100]                  --
│    │    └─Linear: 3-4                  [1, 100]                  10,100
│    │    └─Identity: 3-5                [1, 100]                  --
│    └─FCBlock: 2-4                      --                        (recursive)
│    │    └─LeakyReLU: 3-6               [1, 100]                  --
│    └─FCBlock: 2-5                      [1, 100]                  --
│    │    └─Linear: 3-7                  [1, 100]                  10,100
│    │    └─Identity: 3-8                [1, 100]                  --
│    │    └─LeakyReLU: 3-9               [1, 100]                  --
│    └─Linear: 2-6                       [1, 100]                  10,100
==========================================================================================
Total params: 31,100
Trainable params: 31,100
Non-trainable params: 0
Total mult-adds (M): 0.03
==========================================================================================
Input size (MB): 0.00
Forward/backward pass size (MB): 0.00
Params size (MB): 0.12
Estimated Total Size (MB): 0.13
==========================================================================================

---

source

CoefficientsFC

 CoefficientsFC (n_parallel:int=32, n_biquads:int=2,
                 initial_gain_scale:float=1.0, tahn_c:float=1.0,
                 tanh_d:float=1.0, use_zp_init:bool=False,
                 initial_pole_mag:float=1.5, initial_zero_mag:float=0,
                 **kwargs)

A wrapper around the FC class that outputs the coefficients of a filterbank

Test

n_parallel = 32
n_biquads = 2
n_coeffs = 6
n_batches = 2
coefficients_fc = CoefficientsFC(
    input_size=1007,
    hidden_sizes=[100, 100],
    n_parallel=n_parallel,
    n_biquads=n_biquads,
    initial_gain_scale=1.0,
)

x = torch.randn(n_batches, 1007)
y = coefficients_fc(x)
assert y.shape == torch.Size([n_batches, n_parallel, n_biquads, n_coeffs])