def infer(self, cond_input, implementation):
# cond_input is channels x batch x num_layers x samples
assert(cond_input.size()[0:3:2] == (2*self.R, self.num_layers)), \
"""Inputs are channels x batch x num_layers x samples.
Channels and num_layers should be sizes: {}
But input is: {}""".format((2*self.R, self.num_layers),
cond_input.size()[0:3:2])
batch_size = cond_input.size(1)
sample_count = cond_input.size(3)
cond_input = column_major(cond_input)
samples = torch.cuda.IntTensor(batch_size, sample_count)
nv_wavenet_ext.infer(samples, sample_count, batch_size,
self.embedding_prev, self.embedding_curr,
self.conv_out, self.conv_end, cond_input,
self.num_layers, self.use_embed_tanh,
self.max_dilation, implementation, *self.layers)
return samples
def infer(self, cond_input, implementation):
# cond_input is channels x batch x num_layers x samples
try:
import nv_wavenet_ext
except ImportError as e:
raise ImportError(
f"{str(e)}. Probably you haven't compiled nv_wavenet. "
f"nv_wavenet can be compiled as follows:\n\n"
f"cd {Path(__file__).parent}\n"
f"make\n"
f"python build.py install\n"
)
assert(cond_input.size()[0:3:2] == (2*self.R, self.num_layers)), \
"""Inputs are channels x batch x num_layers x samples.
Channels and num_layers should be sizes: {}
But input is: {}""".format(
(2*self.R, self.num_layers), cond_input.size()[0:3:2])
batch_size = cond_input.size(1)
sample_count = cond_input.size(3)
cond_input = column_major(cond_input)
samples = torch.cuda.IntTensor(batch_size, sample_count)
nv_wavenet_ext.infer(
samples,
sample_count,
batch_size,
self.embedding_prev,
self.embedding_curr,
self.conv_out,
self.conv_end,
cond_input,
self.num_layers,
self.use_embed_tanh,
self.max_dilation,
implementation,
self.layers
)
return samples