of other integer types) in `tf.constant`.
*   Make the `gain` argument of convolutional orthogonal initializers
    (`convolutional_delta_orthogonal`, `convolutional_orthogonal_1D`,
    `convolutional_orthogonal_2D`, `convolutional_orthogonal_3D`) have
    consistent behavior with the `tf.initializers.orthogonal` initializer, i.e.
    scale the output l2-norm by `gain` and NOT by `sqrt(gain)`. (Note that these

FC29          ; mapped                 ; 0639 062C     # 1.1  ARABIC LIGATURE AIN WITH JEEM ISOLATED FORM
FC2A          ; mapped                 ; 0639 0645     # 1.1  ARABIC LIGATURE AIN WITH MEEM ISOLATED FORM
FC2B          ; mapped                 ; 063A 062C     # 1.1  ARABIC LIGATURE GHAIN WITH JEEM ISOLATED FORM
FC2C          ; mapped                 ; 063A 0645     # 1.1  ARABIC LIGATURE GHAIN WITH MEEM ISOLATED FORM

E,MAAA,QDqCF,aC9CA,MAAA,QpBKE,iBAAA,QoBHF,aAAA,QAEA,gBACE,iBAAA,QAGF,yBACE,MAAA,QDqCF,YC9CA,MAAA,QpBKE,iBAAA,QoBHF,aAAA,QAEA,eACE,iBAAA,QAGF,wBACE,MAAA,QCRF,wCACE,KAAO,oBAAA,KAAA,EACP,GAAK,oBAAA,EAAA,GAFP,gCACE,KAAO,oBAAA,KAAA,EACP,GAAK,oBAAA,EAAA,GAIT,UACE,QAAA,YAAA,QAAA,KACA,OAAA,KACA,SAAA,OvCoHI,UAAA,OuClHJ,iBAAA,QhCRE,cAAA,OgCaJ,cACE,QAAA,YAAA,QAAA,KACA,mBAAA,OAAA,eAAA,OACA,cAAA,OAAA,gBAAA,OACA,SAAA,OACA,MAAA,KACA,WAAA,OACA,YAAA,OACA,iBAAA,QvBpBI,WAAA,MAAA,IAAA,KAKF,uCuBOJ,cvBNM,WAAA,MuBkBN,...