l2009312042

when complier the sepcon，occur the following error？

9

src/SeparableConvolution_kernel.cu(66): error: pointer to incomplete class type is not allowed ... src/SeparableConvolution_kernel.cu(170): error: pointer to incomplete class type is not allowed src/SeparableConvolution_kernel.cu(170): error: pointer to incomplete class type is not...

load mpc model error

2

i follow the step with asr/aishell run.sh modify as hkust ,the finetune stage when load pretrain model mpc，it report the following error , the tensorflow verison is 2.3.1 cuda 10.1...

i read the positon encoding code found that ` def call(self, x): """ call function """ seq_len = tf.shape(x)[1] if self.scale: x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32)) x += self.pos_encoding[:, :seq_len, :]...

for example ,in file MMult1.h, AddDot( k, &A( i,0 ), lda, &B( 0,j ), &C( i,j ) ); the lda should be ldb? it is the stride in matix B...

in style_mod function ,why style[:,0] add 1 ?

1

in file ops.py ,the function style_mod in line 372 , def style_mod(x, w): with tf.variable_scope('StyleMod'): units = x.shape[-1] * 2 style = fully_connected(w, units=units, gain=1.0, lrmul=1.0) style = apply_bias(style, lrmul=1.0)...

 the 2nd scale is (0.5 + 0.3 + 1 + 0.7) /4 = 0.625 not 0.5 ?