Suboptimal convergence when compared with TensorFlow model

Thanks for the answer Rodrigo ! I’m not using any constraint or regularizer and biases are also the same. I’ll try to double check everything again, but this is really weird.