deepbox 0.1.0

package/dist/chunk-QERHVCHC.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/nn/index.ts","../src/nn/module/Module.ts","../src/nn/containers/Sequential.ts","../src/nn/layers/activations.ts","../src/nn/layers/dropout.ts","../src/nn/layers/linear.ts","../src/nn/layers/normalization.ts","../src/nn/layers/attention.ts","../src/nn/layers/conv.ts","../src/nn/layers/recurrent.ts","../src/nn/losses/crossEntropy.ts","../src/nn/losses/index.ts"],"names":["__export","ShapeError","DeepboxError","DTypeError","computeStrides","offsetFromFlatIndex","InvalidParameterError","GradTensor","isDevice","IndexError","relu","sigmoid","tanh","leakyRelu","elu","gelu","softmax","normalizeAxis","logSoftmax","softplus","swish","mish","dropout","randn","mulScalar","parameter","zeros","input2d","output2d","output","reshape","dot","transpose","add","isContiguous","dtypeToTypedArrayCtor","getBigIntElement","getNumericElement","Tensor","ones","mean","ensureNumericDType","variance","noGrad","NotFittedError","im2col","getElementAsNumber","stride0","stride1","sum","shapesEqual","sub","pow","tensor","abs","clip","log","mul","neg","sqrt"],"mappings":";;;;;;AAAA,IAAA,UAAA,GAAA;AAAAA,0BAAA,CAAA,UAAA,EAAA;AAAA,EAAA,SAAA,EAAA,MAAA,SAAA;AAAA,EAAA,WAAA,EAAA,MAAA,WAAA;AAAA,EAAA,MAAA,EAAA,MAAA,MAAA;AAAA,EAAA,MAAA,EAAA,MAAA,MAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,GAAA,EAAA,MAAA,GAAA;AAAA,EAAA,IAAA,EAAA,MAAA,IAAA;AAAA,EAAA,GAAA,EAAA,MAAA,GAAA;AAAA,EAAA,IAAA,EAAA,MAAA,IAAA;AAAA,EAAA,SAAA,EAAA,MAAA,SAAA;AAAA,EAAA,SAAA,EAAA,MAAA,SAAA;AAAA,EAAA,MAAA,EAAA,MAAA,MAAA;AAAA,EAAA,UAAA,EAAA,MAAA,UAAA;AAAA,EAAA,SAAA,EAAA,MAAA,SAAA;AAAA,EAAA,IAAA,EAAA,MAAA,IAAA;AAAA,EAAA,MAAA,EAAA,MAAA,MAAA;AAAA,EAAA,kBAAA,EAAA,MAAA,kBAAA;AAAA,EAAA,GAAA,EAAA,MAAA,GAAA;AAAA,EAAA,IAAA,EAAA,MAAA,IAAA;AAAA,EAAA,UAAA,EAAA,MAAA,UAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,QAAA,EAAA,MAAA,QAAA;AAAA,EAAA,KAAA,EAAA,MAAA,KAAA;AAAA,EAAA,IAAA,EAAA,MAAA,IAAA;AAAA,EAAA,uBAAA,EAAA,MAAA,uBAAA;AAAA,EAAA,sBAAA,EAAA,MAAA,sBAAA;AAAA,EAAA,gCAAA,EAAA,MAAA,gCAAA;AAAA,EAAA,gBAAA,EAAA,MAAA,gBAAA;AAAA,EAAA,SAAA,EAAA,MAAA,SAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,QAAA,EAAA,MAAA;AAAA,CAAA,CAAA;;;ACmBA,SAAS,WAAA,CAAY,GAAsB,CAAA,EAA+B;AACxE,EAAA,IAAI,CAAA,CAAE,MAAA,KAAW,CAAA,CAAE,MAAA,EAAQ,OAAO,KAAA;AAClC,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,CAAA,CAAE,QAAQ,CAAA,EAAA,EAAK;AACjC,IAAA,IAAA,CAAK,CAAA,CAAE,CAAC,CAAA,IAAK,CAAA,OAAQ,EAAE,CAAC,CAAA,IAAK,IAAI,OAAO,KAAA;AAAA,EAC1C;AACA,EAAA,OAAO,IAAA;AACT;AAEA,SAAS,aAAA,CAAc,OAA0B,OAAA,EAAyB;AACxE,EAAA,IAAI,IAAA,GAAO,CAAA;AACX,EAAA,KAAA,MAAW,OAAO,KAAA,EAAO;AACvB,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,GAAG,CAAA,IAAK,MAAM,CAAA,EAAG;AACrC,MAAA,MAAM,IAAIC,6BAAW,CAAA,EAAG,OAAO,+BAA+B,MAAA,CAAO,GAAG,CAAC,CAAA,CAAE,CAAA;AAAA,IAC7E;AACA,IAAA,IAAA,IAAQ,GAAA;AAAA,EACV;AACA,EAAA,OAAO,IAAA;AACT;AAEA,SAAS,gBAAgB,CAAA,EAA4C;AACnE,EAAA,MAAM,OAAO,CAAA,CAAE,IAAA;AACf,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,OAAO,KAAK,KAAA,EAAM;AAAA,EACpB;AACA,EAAA,IAAI,gBAAgB,aAAA,EAAe;AACjC,IAAA,OAAO,KAAA,CAAM,KAAK,IAAI,CAAA;AAAA,EACxB;AACA,EAAA,MAAM,GAAA,GAAM,IAAI,KAAA,CAAc,IAAA,CAAK,MAAM,CAAA;AACzC,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,QAAQ,CAAA,EAAA,EAAK;AACpC,IAAA,MAAM,KAAA,GAAQ,KAAK,CAAC,CAAA;AACpB,IAAA,IAAI,UAAU,MAAA,EAAW;AACvB,MAAA,MAAM,IAAIC,+BAAa,kDAAkD,CAAA;AAAA,IAC3E;AACA,IAAA,GAAA,CAAI,CAAC,CAAA,GAAI,KAAA;AAAA,EACX;AACA,EAAA,OAAO,GAAA;AACT;AAEA,SAAS,uBAAA,CACP,IAAA,EACA,IAAA,EACA,KAAA,EACM;AACN,EAAA,MAAM,IAAA,GAAO,cAAc,KAAA,CAAM,KAAA,EAAO,GAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,MAAA,CAAQ,CAAA;AAC/D,EAAA,IAAI,KAAA,CAAM,IAAA,CAAK,MAAA,KAAW,IAAA,EAAM;AAC9B,IAAA,MAAM,IAAID,4BAAA;AAAA,MACR,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,gBAAgB,KAAA,CAAM,IAAA,CAAK,MAAM,CAAA,2BAAA,EAA8B,IAAI,CAAA;AAAA,KACpF;AAAA,EACF;AACF;AAEA,SAAS,wBAAA,CACP,IAAA,EACA,IAAA,EACA,MAAA,EACA,KAAA,EACM;AACN,EAAA,IAAI,CAAC,WAAA,CAAY,MAAA,CAAO,KAAA,EAAO,KAAA,CAAM,KAAK,CAAA,EAAG;AAC3C,IAAA,MAAM,IAAIA,4BAAA;AAAA,MACR,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,8BAA8B,MAAA,CAAO,KAAA,CAAM,IAAA,CAAK,IAAI,CAAC,CAAA,QAAA,EAAW,KAAA,CAAM,KAAA,CAAM,IAAA,CAAK,IAAI,CAAC,CAAA,CAAA;AAAA,KACvG;AAAA,EACF;AACA,EAAA,IAAI,MAAA,CAAO,KAAA,KAAU,KAAA,CAAM,KAAA,EAAO;AAChC,IAAA,MAAM,IAAIE,4BAAA;AAAA,MACR,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,6BAA6B,MAAA,CAAO,KAAK,CAAA,MAAA,EAAS,KAAA,CAAM,KAAK,CAAA;AAAA,KAC9E;AAAA,EACF;AAEA,EAAA,MAAM,IAAA,GAAO,cAAc,KAAA,CAAM,KAAA,EAAO,GAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,MAAA,CAAQ,CAAA;AAC/D,EAAA,MAAM,cAAA,GAAiBC,gCAAA,CAAe,MAAA,CAAO,KAAK,CAAA;AAClD,EAAA,MAAM,OAAO,MAAA,CAAO,IAAA;AAEpB,EAAA,IAAI,MAAA,CAAO,UAAU,QAAA,EAAU;AAC7B,IAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACxB,MAAA,MAAM,IAAID,4BAAA,CAAW,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,qBAAA,CAAuB,CAAA;AAAA,IAC7D;AACA,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,MAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,IAAA,CAAK,CAAC,CAAA;AAC1B,MAAA,IAAI,OAAO,UAAU,QAAA,EAAU;AAC7B,QAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,oBAAA,CAAsB,CAAA;AAAA,MAC5D;AACA,MAAA,MAAM,SAASE,qCAAA,CAAoB,CAAA,EAAG,gBAAgB,MAAA,CAAO,OAAA,EAAS,OAAO,MAAM,CAAA;AACnF,MAAA,IAAA,CAAK,MAAM,CAAA,GAAI,KAAA;AAAA,IACjB;AACA,IAAA;AAAA,EACF;AAEA,EAAA,IAAI,gBAAgB,aAAA,EAAe;AACjC,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,MAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,IAAA,CAAK,CAAC,CAAA;AAC1B,MAAA,IAAI,OAAO,UAAU,QAAA,EAAU;AAC7B,QAAA,MAAM,IAAIF,4BAAA,CAAW,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,oBAAA,CAAsB,CAAA;AAAA,MAC5D;AACA,MAAA,MAAM,SAASE,qCAAA,CAAoB,CAAA,EAAG,gBAAgB,MAAA,CAAO,OAAA,EAAS,OAAO,MAAM,CAAA;AACnF,MAAA,IAAA,CAAK,MAAM,CAAA,GAAI,KAAA;AAAA,IACjB;AACA,IAAA;AAAA,EACF;AAEA,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,MAAM,IAAIF,4BAAA,CAAW,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,sBAAA,CAAwB,CAAA;AAAA,EAC9D;AAEA,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,IAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,IAAA,CAAK,CAAC,CAAA;AAC1B,IAAA,IAAI,OAAO,UAAU,QAAA,EAAU;AAC7B,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,EAAG,IAAI,CAAA,CAAA,EAAI,IAAI,CAAA,qBAAA,CAAuB,CAAA;AAAA,IAC7D;AACA,IAAA,MAAM,SAASE,qCAAA,CAAoB,CAAA,EAAG,gBAAgB,MAAA,CAAO,OAAA,EAAS,OAAO,MAAM,CAAA;AACnF,IAAA,IAAA,CAAK,MAAM,CAAA,GAAI,KAAA;AAAA,EACjB;AACF;AAmEO,IAAe,MAAA,GAAf,MAAe,OAAA,CAAO;AAAA;AAAA,EAEjB,QAAA,uBAAoC,GAAA,EAAI;AAAA;AAAA,EAGxC,WAAA,uBAA2C,GAAA,EAAI;AAAA;AAAA,EAG/C,QAAA,uBAAoC,GAAA,EAAI;AAAA;AAAA,EAGxC,SAAA,GAAY,IAAA;AAAA;AAAA,EAGd,gBAAA,uBAAoD,GAAA,EAAI;AAAA;AAAA,EAExD,aAAA,uBAA8C,GAAA,EAAI;AAAA;AAAA,EAElD,WAAA,GAAc,CAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EA6BtB,QAAQ,MAAA,EAAgC;AACtC,IAAA,IAAI,SAAA,GAAY,MAAA;AAChB,IAAA,KAAA,MAAW,IAAA,IAAQ,IAAA,CAAK,gBAAA,CAAiB,MAAA,EAAO,EAAG;AACjD,MAAA,MAAM,MAAA,GAAS,IAAA,CAAK,IAAA,EAAM,SAAS,CAAA;AACnC,MAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,MAAM,CAAA,EAAG;AACzB,QAAA,SAAA,GAAY,MAAA;AAAA,MACd;AAAA,IACF;AACA,IAAA,IAAI,MAAA,GAAS,IAAA,CAAK,OAAA,CAAQ,GAAG,SAAS,CAAA;AACtC,IAAA,KAAA,MAAW,IAAA,IAAQ,IAAA,CAAK,aAAA,CAAc,MAAA,EAAO,EAAG;AAC9C,MAAA,MAAM,MAAA,GAAS,IAAA,CAAK,IAAA,EAAM,SAAA,EAAW,MAAM,CAAA;AAC3C,MAAA,IAAI,WAAW,MAAA,EAAW;AACxB,QAAA,MAAA,GAAS,MAAA;AAAA,MACX;AAAA,IACF;AACA,IAAA,OAAO,MAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQU,cAAA,CAAe,MAAc,MAAA,EAAsB;AAE3D,IAAA,IAAA,CAAK,QAAA,CAAS,GAAA,CAAI,IAAA,EAAM,MAAM,CAAA;AAAA,EAChC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAWU,iBAAA,CAAkB,MAAc,KAAA,EAAyB;AAEjE,IAAA,IAAA,CAAK,WAAA,CAAY,GAAA,CAAI,IAAA,EAAM,KAAK,CAAA;AAAA,EAClC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUU,cAAA,CAAe,MAAc,MAAA,EAAsB;AAG3D,IAAA,IAAA,CAAK,QAAA,CAAS,GAAA,CAAI,IAAA,EAAM,MAAM,CAAA;AAAA,EAChC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAcA,CAAC,UAAA,CAAW,OAAA,GAAU,IAAA,EAA6B;AAEjD,IAAA,KAAA,MAAW,KAAA,IAAS,IAAA,CAAK,WAAA,CAAY,MAAA,EAAO,EAAG;AAC7C,MAAA,MAAM,KAAA;AAAA,IACR;AAIA,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,QAAA,CAAS,MAAA,EAAO,EAAG;AAC3C,QAAA,OAAO,MAAA,CAAO,WAAW,IAAI,CAAA;AAAA,MAC/B;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASA,CAAC,eAAA,CAAgB,MAAA,GAAS,EAAA,EAAI,UAAU,IAAA,EAAuC;AAE7E,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,KAAK,KAAK,IAAA,CAAK,WAAA,CAAY,SAAQ,EAAG;AAEtD,MAAA,MAAM,WAAW,MAAA,GAAS,CAAA,EAAG,MAAM,CAAA,CAAA,EAAI,IAAI,CAAA,CAAA,GAAK,IAAA;AAChD,MAAA,MAAM,CAAC,UAAU,KAAK,CAAA;AAAA,IACxB;AAGA,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,KAAA,MAAW,CAAC,UAAA,EAAY,MAAM,KAAK,IAAA,CAAK,QAAA,CAAS,SAAQ,EAAG;AAE1D,QAAA,MAAM,aAAa,MAAA,GAAS,CAAA,EAAG,MAAM,CAAA,CAAA,EAAI,UAAU,CAAA,CAAA,GAAK,UAAA;AACxD,QAAA,OAAO,MAAA,CAAO,eAAA,CAAgB,UAAA,EAAY,IAAI,CAAA;AAAA,MAChD;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,CAAC,OAAA,CAAQ,OAAA,GAAU,IAAA,EAAyB;AAE1C,IAAA,MAAM,IAAA;AAGN,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,QAAA,CAAS,MAAA,EAAO,EAAG;AAC3C,QAAA,OAAO,MAAA,CAAO,QAAQ,IAAI,CAAA;AAAA,MAC5B;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASA,CAAC,YAAA,CAAa,MAAA,GAAS,EAAA,EAAI,UAAU,IAAA,EAAmC;AAEtE,IAAA,MAAM,CAAC,QAAQ,IAAI,CAAA;AAGnB,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,KAAA,MAAW,CAAC,IAAA,EAAM,MAAM,KAAK,IAAA,CAAK,QAAA,CAAS,SAAQ,EAAG;AAEpD,QAAA,MAAM,WAAW,MAAA,GAAS,CAAA,EAAG,MAAM,CAAA,CAAA,EAAI,IAAI,CAAA,CAAA,GAAK,IAAA;AAChD,QAAA,OAAO,MAAA,CAAO,YAAA,CAAa,QAAA,EAAU,IAAI,CAAA;AAAA,MAC3C;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUA,KAAA,CAAM,OAAO,IAAA,EAAY;AAEvB,IAAA,IAAA,CAAK,SAAA,GAAY,IAAA;AAIjB,IAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,QAAA,CAAS,MAAA,EAAO,EAAG;AAC3C,MAAA,MAAA,CAAO,MAAM,IAAI,CAAA;AAAA,IACnB;AAGA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASA,IAAA,GAAa;AACX,IAAA,OAAO,IAAA,CAAK,MAAM,KAAK,CAAA;AAAA,EACzB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,IAAI,QAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,SAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAmBA,QAAA,GAAiB;AAGf,IAAA,KAAA,MAAW,KAAA,IAAS,IAAA,CAAK,UAAA,EAAW,EAAG;AAErC,MAAA,KAAA,CAAM,QAAA,EAAS;AAAA,IACjB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,CAAC,OAAA,CAAQ,OAAA,GAAU,IAAA,EAAyB;AAC1C,IAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,QAAA,CAAS,MAAA,EAAO,EAAG;AAC3C,MAAA,MAAM,MAAA;AAAA,IACR;AACA,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,QAAA,CAAS,MAAA,EAAO,EAAG;AAC3C,QAAA,OAAO,MAAA,CAAO,QAAQ,IAAI,CAAA;AAAA,MAC5B;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,CAAC,YAAA,CAAa,MAAA,GAAS,EAAA,EAAI,UAAU,IAAA,EAAmC;AACtE,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,MAAM,KAAK,IAAA,CAAK,QAAA,CAAS,SAAQ,EAAG;AACpD,MAAA,MAAM,WAAW,MAAA,GAAS,CAAA,EAAG,MAAM,CAAA,CAAA,EAAI,IAAI,CAAA,CAAA,GAAK,IAAA;AAChD,MAAA,MAAM,CAAC,UAAU,MAAM,CAAA;AAAA,IACzB;AACA,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,KAAA,MAAW,CAAC,UAAA,EAAY,MAAM,KAAK,IAAA,CAAK,QAAA,CAAS,SAAQ,EAAG;AAC1D,QAAA,MAAM,aAAa,MAAA,GAAS,CAAA,EAAG,MAAM,CAAA,CAAA,EAAI,UAAU,CAAA,CAAA,GAAK,UAAA;AACxD,QAAA,OAAO,MAAA,CAAO,YAAA,CAAa,UAAA,EAAY,IAAI,CAAA;AAAA,MAC7C;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAsBA,gBAAA,CAAiB,KAAA,EAAkB,OAAA,GAAU,IAAA,EAAY;AACvD,IAAA,IAAA,CAAK,uBAAA,CAAwB,KAAA,EAAO,KAAA,EAAO,OAAO,CAAA;AAAA,EACpD;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAsBA,kBAAA,CAAmB,KAAA,EAAkB,OAAA,GAAU,IAAA,EAAY;AACzD,IAAA,IAAA,CAAK,uBAAA,CAAwB,KAAA,EAAO,IAAA,EAAM,OAAO,CAAA;AAAA,EACnD;AAAA,EAEQ,uBAAA,CACN,KAAA,EACA,YAAA,EACA,OAAA,EACM;AACN,IAAA,MAAM,gBAAgB,KAAA,KAAU,MAAA;AAChC,IAAA,MAAM,WAAA,GACJ,KAAA,IAAS,KAAA,CAAM,IAAA,CAAK,KAAK,eAAA,CAAgB,EAAA,EAAI,OAAO,CAAC,EAAE,GAAA,CAAI,CAAC,CAAC,IAAI,MAAM,IAAI,CAAA;AAC7E,IAAA,KAAA,MAAW,QAAQ,WAAA,EAAa;AAC9B,MAAA,MAAM,QAAA,GAAW,IAAA,CAAK,oBAAA,CAAqB,IAAI,CAAA;AAC/C,MAAA,IAAI,CAAC,QAAA,EAAU;AACb,QAAA,IAAI,aAAA,EAAe;AACjB,UAAA,MAAM,IAAIC,uCAAA,CAAsB,CAAA,wBAAA,EAA2B,IAAI,CAAA,CAAA,EAAI,SAAS,IAAI,CAAA;AAAA,QAClF;AACA,QAAA;AAAA,MACF;AACA,MAAA,MAAM,EAAE,MAAA,EAAQ,SAAA,EAAU,GAAI,QAAA;AAC9B,MAAA,MAAM,KAAA,GAAQ,MAAA,CAAO,WAAA,CAAY,GAAA,CAAI,SAAS,CAAA;AAC9C,MAAA,IAAI,CAAC,KAAA,EAAO;AACV,QAAA,IAAI,aAAA,EAAe;AACjB,UAAA,MAAM,IAAIA,uCAAA,CAAsB,CAAA,wBAAA,EAA2B,IAAI,CAAA,CAAA,EAAI,SAAS,IAAI,CAAA;AAAA,QAClF;AACA,QAAA;AAAA,MACF;AAEA,MAAA,MAAM,YAAYC,4BAAA,CAAW,UAAA,CAAW,MAAM,MAAA,EAAQ,EAAE,cAAc,CAAA;AACtE,MAAA,MAAA,CAAO,WAAA,CAAY,GAAA,CAAI,SAAA,EAAW,SAAS,CAAA;AAC3C,MAAA,KAAA,MAAW,CAAC,GAAA,EAAK,KAAK,KAAK,MAAA,CAAO,OAAA,CAAQ,MAAM,CAAA,EAAG;AACjD,QAAA,IAAI,UAAU,KAAA,EAAO;AACnB,UAAA,OAAA,CAAQ,GAAA,CAAI,MAAA,EAAQ,GAAA,EAAK,SAAS,CAAA;AAAA,QACpC;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,qBAAqB,QAAA,EAAgE;AAC3F,IAAA,MAAM,KAAA,GAAQ,QAAA,CAAS,KAAA,CAAM,GAAG,CAAA;AAChC,IAAA,IAAI,MAAA,GAAiB,IAAA;AACrB,IAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,KAAA,CAAM,MAAA,GAAS,GAAG,CAAA,EAAA,EAAK;AACzC,MAAA,MAAM,IAAA,GAAO,KAAA,CAAM,CAAC,CAAA,IAAK,EAAA;AACzB,MAAA,MAAM,KAAA,GAAQ,MAAA,CAAO,QAAA,CAAS,GAAA,CAAI,IAAI,CAAA;AACtC,MAAA,IAAI,CAAC,OAAO,OAAO,IAAA;AACnB,MAAA,MAAA,GAAS,KAAA;AAAA,IACX;AACA,IAAA,MAAM,SAAA,GAAY,KAAA,CAAM,KAAA,CAAM,MAAA,GAAS,CAAC,CAAA,IAAK,EAAA;AAC7C,IAAA,OAAO,EAAE,QAAQ,SAAA,EAAU;AAAA,EAC7B;AAAA,EAEA,OAAe,uBAAA,CAAwB,MAAA,EAAgB,MAAA,EAAsB;AAC3E,IAAA,IAAI,CAAC,OAAA,CAAQ,GAAA,CAAI,MAAA,EAAQ,QAAA,EAAU,MAAM,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIL,+BAAa,yCAAyC,CAAA;AAAA,IAClE;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,SAAA,GAGE;AACA,IAAA,MAAM,aAAyC,EAAC;AAChD,IAAA,MAAM,UAAsC,EAAC;AAE7C,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,KAAK,CAAA,IAAK,IAAA,CAAK,iBAAgB,EAAG;AAClD,MAAA,MAAM,IAAI,KAAA,CAAM,MAAA;AAChB,MAAA,MAAM,IAAA,GAAO,gBAAgB,CAAC,CAAA;AAC9B,MAAA,UAAA,CAAW,IAAI,CAAA,GAAI;AAAA,QACjB,IAAA;AAAA,QACA,KAAA,EAAO,CAAC,GAAG,CAAA,CAAE,KAAK,CAAA;AAAA,QAClB,OAAO,CAAA,CAAE;AAAA,OACX;AAAA,IACF;AAEA,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,MAAM,CAAA,IAAK,IAAA,CAAK,cAAa,EAAG;AAChD,MAAA,MAAM,IAAA,GAAO,gBAAgB,MAAM,CAAA;AACnC,MAAA,OAAA,CAAQ,IAAI,CAAA,GAAI;AAAA,QACd,IAAA;AAAA,QACA,KAAA,EAAO,CAAC,GAAG,MAAA,CAAO,KAAK,CAAA;AAAA,QACvB,OAAO,MAAA,CAAO;AAAA,OAChB;AAAA,IACF;AAEA,IAAA,OAAO,EAAE,YAAY,OAAA,EAAQ;AAAA,EAC/B;AAAA;AAAA;AAAA;AAAA,EAKA,cAAc,SAAA,EAGL;AACP,IAAA,MAAM,UAAA,GAAa,SAAA,CAAU,UAAA,IAAc,EAAC;AAC5C,IAAA,MAAM,OAAA,GAAU,SAAA,CAAU,OAAA,IAAW,EAAC;AAEtC,IAAA,MAAM,WAAA,GAAc,IAAI,GAAA,CAAI,IAAA,CAAK,iBAAiB,CAAA;AAClD,IAAA,MAAM,UAAA,GAAa,IAAI,GAAA,CAAI,IAAA,CAAK,cAAc,CAAA;AAE9C,IAAA,KAAA,MAAW,IAAA,IAAQ,WAAA,CAAY,IAAA,EAAK,EAAG;AACrC,MAAA,IAAI,EAAE,QAAQ,UAAA,CAAA,EAAa;AACzB,QAAA,MAAM,IAAII,uCAAA,CAAsB,CAAA,mBAAA,EAAsB,IAAI,CAAA,CAAA,EAAI,wBAAwB,IAAI,CAAA;AAAA,MAC5F;AAAA,IACF;AAEA,IAAA,KAAA,MAAW,IAAA,IAAQ,UAAA,CAAW,IAAA,EAAK,EAAG;AACpC,MAAA,IAAI,EAAE,QAAQ,OAAA,CAAA,EAAU;AACtB,QAAA,MAAM,IAAIA,uCAAA,CAAsB,CAAA,gBAAA,EAAmB,IAAI,CAAA,CAAA,EAAI,qBAAqB,IAAI,CAAA;AAAA,MACtF;AAAA,IACF;AAEA,IAAA,KAAA,MAAW,IAAA,IAAQ,MAAA,CAAO,IAAA,CAAK,UAAU,CAAA,EAAG;AAC1C,MAAA,IAAI,CAAC,WAAA,CAAY,GAAA,CAAI,IAAI,CAAA,EAAG;AAC1B,QAAA,MAAM,IAAIA,uCAAA;AAAA,UACR,yBAAyB,IAAI,CAAA,CAAA;AAAA,UAC7B,sBAAA;AAAA,UACA;AAAA,SACF;AAAA,MACF;AAAA,IACF;AAEA,IAAA,KAAA,MAAW,IAAA,IAAQ,MAAA,CAAO,IAAA,CAAK,OAAO,CAAA,EAAG;AACvC,MAAA,IAAI,CAAC,UAAA,CAAW,GAAA,CAAI,IAAI,CAAA,EAAG;AACzB,QAAA,MAAM,IAAIA,uCAAA,CAAsB,CAAA,mBAAA,EAAsB,IAAI,CAAA,CAAA,EAAI,qBAAqB,IAAI,CAAA;AAAA,MACzF;AAAA,IACF;AAEA,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,KAAK,KAAK,MAAA,CAAO,OAAA,CAAQ,UAAU,CAAA,EAAG;AACtD,MAAA,MAAM,KAAA,GAAQ,WAAA,CAAY,GAAA,CAAI,IAAI,CAAA;AAClC,MAAA,IAAI,CAAC,KAAA,EAAO;AACZ,MAAA,uBAAA,CAAwB,IAAA,EAAM,aAAa,KAAK,CAAA;AAChD,MAAA,wBAAA,CAAyB,IAAA,EAAM,WAAA,EAAa,KAAA,CAAM,MAAA,EAAQ,KAAK,CAAA;AAAA,IACjE;AAEA,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,KAAK,KAAK,MAAA,CAAO,OAAA,CAAQ,OAAO,CAAA,EAAG;AACnD,MAAA,MAAM,MAAA,GAAS,UAAA,CAAW,GAAA,CAAI,IAAI,CAAA;AAClC,MAAA,IAAI,CAAC,MAAA,EAAQ;AACb,MAAA,uBAAA,CAAwB,IAAA,EAAM,UAAU,KAAK,CAAA;AAC7C,MAAA,wBAAA,CAAyB,IAAA,EAAM,QAAA,EAAU,MAAA,EAAQ,KAAK,CAAA;AAAA,IACxD;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAsBA,GAAG,MAAA,EAAsB;AACvB,IAAA,IAAI,CAACE,0BAAA,CAAS,MAAM,CAAA,EAAG;AACrB,MAAA,MAAM,IAAIF,uCAAA,CAAsB,0CAAA,EAA4C,QAAA,EAAU,MAAM,CAAA;AAAA,IAC9F;AAEA,IAAA,KAAA,MAAW,KAAA,IAAS,IAAA,CAAK,UAAA,EAAW,EAAG;AACrC,MAAA,OAAA,CAAO,uBAAA,CAAwB,KAAA,CAAM,MAAA,EAAQ,MAAM,CAAA;AAAA,IACrD;AACA,IAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,OAAA,EAAQ,EAAG;AACnC,MAAA,OAAA,CAAO,uBAAA,CAAwB,QAAQ,MAAM,CAAA;AAAA,IAC/C;AACA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA,EAKA,MAAM,EAAA,EAAoC;AACxC,IAAA,KAAA,MAAW,MAAA,IAAU,IAAA,CAAK,OAAA,EAAQ,EAAG;AACnC,MAAA,EAAA,CAAG,MAAM,CAAA;AAAA,IACX;AACA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA,EAKA,uBAAuB,IAAA,EAAkC;AACvD,IAAA,MAAM,SAAS,IAAA,CAAK,WAAA,EAAA;AACpB,IAAA,IAAA,CAAK,gBAAA,CAAiB,GAAA,CAAI,MAAA,EAAQ,IAAI,CAAA;AACtC,IAAA,OAAO,MAAM;AACX,MAAA,IAAA,CAAK,gBAAA,CAAiB,OAAO,MAAM,CAAA;AAAA,IACrC,CAAA;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,oBAAoB,IAAA,EAA+B;AACjD,IAAA,MAAM,SAAS,IAAA,CAAK,WAAA,EAAA;AACpB,IAAA,IAAA,CAAK,aAAA,CAAc,GAAA,CAAI,MAAA,EAAQ,IAAI,CAAA;AACnC,IAAA,OAAO,MAAM;AACX,MAAA,IAAA,CAAK,aAAA,CAAc,OAAO,MAAM,CAAA;AAAA,IAClC,CAAA;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,QAAA,GAAmB;AACjB,IAAA,MAAM,QAAQ,CAAC,CAAA,EAAG,IAAA,CAAK,WAAA,CAAY,IAAI,CAAA,CAAA,CAAG,CAAA;AAG1C,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,MAAM,KAAK,IAAA,CAAK,QAAA,CAAS,SAAQ,EAAG;AAEpD,MAAA,MAAM,UAAA,GAAa,MAAA,CAAO,QAAA,EAAS,CAAE,MAAM,IAAI,CAAA;AAE/C,MAAA,MAAM,SAAA,GAAY,UAAA,CAAW,GAAA,CAAI,CAAC,MAAM,CAAA,KAAO,CAAA,KAAM,CAAA,GAAI,IAAA,GAAO,CAAA,EAAA,EAAK,IAAI,CAAA,CAAG,CAAA,CAAE,KAAK,IAAI,CAAA;AAEvF,MAAA,KAAA,CAAM,IAAA,CAAK,CAAA,GAAA,EAAM,IAAI,CAAA,GAAA,EAAM,SAAS,CAAA,CAAE,CAAA;AAAA,IACxC;AAEA,IAAA,KAAA,CAAM,KAAK,GAAG,CAAA;AACd,IAAA,OAAO,KAAA,CAAM,KAAK,IAAI,CAAA;AAAA,EACxB;AACF;;;AC5sBO,IAAM,UAAA,GAAN,cAAyB,MAAA,CAAO;AAAA;AAAA,EAEpB,MAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASjB,eAAe,MAAA,EAAkB;AAC/B,IAAA,KAAA,EAAM;AAGN,IAAA,IAAI,MAAA,CAAO,WAAW,CAAA,EAAG;AACvB,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,wCAAA;AAAA,QACA,QAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAAA,IACF;AAGA,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AAId,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,MAAA,CAAO,QAAQ,CAAA,EAAA,EAAK;AACtC,MAAA,MAAM,KAAA,GAAQ,OAAO,CAAC,CAAA;AACtB,MAAA,IAAI,CAAC,KAAA,EAAO;AACV,QAAA,MAAM,IAAIJ,8BAAA,CAAa,CAAA,eAAA,EAAkB,CAAC,CAAA,aAAA,CAAe,CAAA;AAAA,MAC3D;AACA,MAAA,IAAA,CAAK,cAAA,CAAe,MAAA,CAAO,CAAC,CAAA,EAAG,KAAK,CAAA;AAAA,IACtC;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAYA,WAAW,MAAA,EAAgC;AACzC,IAAA,IAAI,MAAA,CAAO,WAAW,CAAA,EAAG;AACvB,MAAA,MAAM,IAAII,uCAAA;AAAA,QACR,kDAAA;AAAA,QACA,QAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAAA,IACF;AACA,IAAA,MAAM,KAAA,GAAQ,OAAO,CAAC,CAAA;AACtB,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,kDAAA;AAAA,QACA,OAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AAEA,IAAA,IAAI,MAAA,GAAS,KAAA;AAIb,IAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,MAAA,CAAO,QAAQ,CAAA,EAAA,EAAK;AAC3C,MAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC3B,MAAA,IAAI,CAAC,KAAA,EAAO;AACV,QAAA,MAAM,IAAIJ,8BAAA,CAAa,CAAA,eAAA,EAAkB,CAAC,CAAA,aAAA,CAAe,CAAA;AAAA,MAC3D;AAIA,MAAA,MAAM,MAAA,GAAS,KAAA,CAAM,IAAA,CAAK,MAAM,CAAA;AAChC,MAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,MAAM,CAAA,EAAG;AACzB,QAAA,MAAM,IAAII,uCAAA;AAAA,UACR,0EAA0E,CAAC,CAAA,CAAA,CAAA;AAAA,UAC3E,OAAA;AAAA,UACA;AAAA,SACF;AAAA,MACF;AACA,MAAA,MAAA,GAAS,MAAA;AAAA,IACX;AAGA,IAAA,OAAO,MAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUA,SAAS,KAAA,EAAuB;AAE9B,IAAA,IAAI,KAAA,GAAQ,CAAA,IAAK,KAAA,IAAS,IAAA,CAAK,OAAO,MAAA,EAAQ;AAC5C,MAAA,MAAM,IAAIG,6BAAW,CAAA,YAAA,EAAe,KAAK,sBAAsB,IAAA,CAAK,MAAA,CAAO,MAAM,CAAA,CAAA,CAAA,EAAK;AAAA,QACpF,KAAA;AAAA,QACA,YAAY,CAAC,CAAA,EAAG,IAAA,CAAK,MAAA,CAAO,SAAS,CAAC;AAAA,OACvC,CAAA;AAAA,IACH;AAEA,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,MAAA,CAAO,KAAK,CAAA;AAC/B,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAIP,8BAAA,CAAa,CAAA,eAAA,EAAkB,KAAK,CAAA,aAAA,CAAe,CAAA;AAAA,IAC/D;AAEA,IAAA,OAAO,KAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA,EAKA,IAAI,MAAA,GAAiB;AACnB,IAAA,OAAO,KAAK,MAAA,CAAO,MAAA;AAAA,EACrB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOS,QAAA,GAAmB;AAE1B,IAAA,MAAM,KAAA,GAAQ,CAAC,aAAa,CAAA;AAG5B,IAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,MAAA,CAAO,QAAQ,CAAA,EAAA,EAAK;AAC3C,MAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC3B,MAAA,IAAI,CAAC,KAAA,EAAO;AAGZ,MAAA,MAAM,UAAA,GAAa,KAAA,CAAM,QAAA,EAAS,CAAE,MAAM,IAAI,CAAA;AAC9C,MAAA,MAAM,QAAA,GAAW,UAAA,CAAW,GAAA,CAAI,CAAC,MAAM,GAAA,KAAS,GAAA,KAAQ,CAAA,GAAI,IAAA,GAAO,CAAA,EAAA,EAAK,IAAI,CAAA,CAAG,CAAA,CAAE,KAAK,IAAI,CAAA;AAG1F,MAAA,KAAA,CAAM,IAAA,CAAK,CAAA,GAAA,EAAM,CAAC,CAAA,GAAA,EAAM,QAAQ,CAAA,CAAE,CAAA;AAAA,IACpC;AAEA,IAAA,KAAA,CAAM,KAAK,GAAG,CAAA;AACd,IAAA,OAAO,KAAA,CAAM,KAAK,IAAI,CAAA;AAAA,EACxB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,EAAE,MAAA,CAAO,QAAQ,CAAA,GAAsB;AACrC,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,MAAA,EAAQ;AAC/B,MAAA,MAAM,KAAA;AAAA,IACR;AAAA,EACF;AACF;;;ACzLO,IAAM,IAAA,GAAN,cAAmB,MAAA,CAAO;AAAA,EAG/B,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,KAAA,YAAiBK,4BAAA,EAAY,OAAO,KAAA,CAAM,IAAA,EAAK;AACnD,IAAA,OAAOG,uBAAO,KAAK,CAAA;AAAA,EACrB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,QAAA;AAAA,EACT;AACF;AASO,IAAM,OAAA,GAAN,cAAsB,MAAA,CAAO;AAAA,EAGlC,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,KAAA,YAAiBH,4BAAA,EAAY,OAAO,KAAA,CAAM,OAAA,EAAQ;AACtD,IAAA,OAAOI,0BAAU,KAAK,CAAA;AAAA,EACxB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,WAAA;AAAA,EACT;AACF;AASO,IAAM,IAAA,GAAN,cAAmB,MAAA,CAAO;AAAA,EAG/B,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,KAAA,YAAiBJ,4BAAA,EAAY,OAAO,KAAA,CAAM,IAAA,EAAK;AACnD,IAAA,OAAOK,uBAAO,KAAK,CAAA;AAAA,EACrB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,QAAA;AAAA,EACT;AACF;AASO,IAAM,SAAA,GAAN,cAAwB,MAAA,CAAO;AAAA,EACnB,KAAA;AAAA,EAEjB,WAAA,CAAY,QAAQ,IAAA,EAAM;AACxB,IAAA,KAAA,EAAM;AACN,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AAAA,EACf;AAAA,EAIA,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBL,4BAAA,EAAY,OAAO,KAAA,CAAM,SAAA,CAAU,KAAK,KAAK,CAAA;AAClE,IAAA,OAAOM,2BAAA,CAAY,KAAA,EAAO,IAAA,CAAK,KAAK,CAAA;AAAA,EACtC;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,gBAAA,EAAmB,KAAK,KAAK,CAAA,CAAA,CAAA;AAAA,EACtC;AACF;AASO,IAAM,GAAA,GAAN,cAAkB,MAAA,CAAO;AAAA,EACb,KAAA;AAAA,EAEjB,WAAA,CAAY,QAAQ,CAAA,EAAK;AACvB,IAAA,KAAA,EAAM;AAGN,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AAAA,EACf;AAAA,EAIA,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBN,4BAAA,EAAY,OAAO,KAAA,CAAM,GAAA,CAAI,KAAK,KAAK,CAAA;AAC5D,IAAA,OAAOO,qBAAA,CAAM,KAAA,EAAO,IAAA,CAAK,KAAK,CAAA;AAAA,EAChC;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,UAAA,EAAa,KAAK,KAAK,CAAA,CAAA,CAAA;AAAA,EAChC;AACF;AASO,IAAM,IAAA,GAAN,cAAmB,MAAA,CAAO;AAAA,EAG/B,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,KAAA,YAAiBP,4BAAA,EAAY,OAAO,KAAA,CAAM,IAAA,EAAK;AACnD,IAAA,OAAOQ,uBAAO,KAAK,CAAA;AAAA,EACrB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,QAAA;AAAA,EACT;AACF;AASO,IAAM,OAAA,GAAN,cAAsB,MAAA,CAAO;AAAA,EACjB,IAAA;AAAA,EAEjB,WAAA,CAAY,OAAa,EAAA,EAAI;AAC3B,IAAA,KAAA,EAAM;AAGN,IAAA,IAAA,CAAK,IAAA,GAAO,IAAA;AAAA,EACd;AAAA,EAIA,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBR,4BAAA,EAAY;AAC/B,MAAA,OAAOS,0BAAAA,CAAY,OAAOC,+BAAA,CAAc,IAAA,CAAK,MAAM,KAAA,CAAM,MAAA,CAAO,IAAI,CAAC,CAAA;AAAA,IACvE;AACA,IAAA,OAAOD,yBAAA,CAAU,KAAA,EAAO,IAAA,CAAK,IAAI,CAAA;AAAA,EACnC;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,aAAA,EAAgB,KAAK,IAAI,CAAA,CAAA,CAAA;AAAA,EAClC;AACF;AASO,IAAM,UAAA,GAAN,cAAyB,MAAA,CAAO;AAAA,EACpB,IAAA;AAAA,EAEjB,WAAA,CAAY,OAAa,EAAA,EAAI;AAC3B,IAAA,KAAA,EAAM;AAGN,IAAA,IAAA,CAAK,IAAA,GAAO,IAAA;AAAA,EACd;AAAA,EAIA,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBT,4BAAA,EAAY;AAC/B,MAAA,OAAOW,6BAAAA,CAAe,OAAOD,+BAAA,CAAc,IAAA,CAAK,MAAM,KAAA,CAAM,MAAA,CAAO,IAAI,CAAC,CAAA;AAAA,IAC1E;AACA,IAAA,OAAOC,4BAAA,CAAa,KAAA,EAAO,IAAA,CAAK,IAAI,CAAA;AAAA,EACtC;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,gBAAA,EAAmB,KAAK,IAAI,CAAA,CAAA,CAAA;AAAA,EACrC;AACF;AASO,IAAM,QAAA,GAAN,cAAuB,MAAA,CAAO;AAAA,EAGnC,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBX,4BAAA,EAAY;AAC/B,MAAA,OAAOA,4BAAA,CAAW,UAAA,CAAWY,0BAAA,CAAW,KAAA,CAAM,MAAM,CAAA,EAAG;AAAA,QACrD,YAAA,EAAc;AAAA,OACf,CAAA;AAAA,IACH;AACA,IAAA,OAAOA,2BAAW,KAAK,CAAA;AAAA,EACzB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,YAAA;AAAA,EACT;AACF;AASO,IAAM,KAAA,GAAN,cAAoB,MAAA,CAAO;AAAA,EAGhC,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBZ,4BAAA,EAAY;AAC/B,MAAA,OAAOA,4BAAA,CAAW,UAAA,CAAWa,uBAAA,CAAQ,KAAA,CAAM,MAAM,CAAA,EAAG;AAAA,QAClD,YAAA,EAAc;AAAA,OACf,CAAA;AAAA,IACH;AACA,IAAA,OAAOA,wBAAQ,KAAK,CAAA;AAAA,EACtB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,SAAA;AAAA,EACT;AACF;AASO,IAAM,IAAA,GAAN,cAAmB,MAAA,CAAO;AAAA,EAG/B,QAAQ,KAAA,EAAiD;AACvD,IAAA,IAAI,iBAAiBb,4BAAA,EAAY;AAC/B,MAAA,OAAOA,4BAAA,CAAW,UAAA,CAAWc,sBAAA,CAAO,KAAA,CAAM,MAAM,CAAA,EAAG;AAAA,QACjD,YAAA,EAAc;AAAA,OACf,CAAA;AAAA,IACH;AACA,IAAA,OAAOA,uBAAO,KAAK,CAAA;AAAA,EACrB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,QAAA;AAAA,EACT;AACF;;;AC3OO,IAAM,OAAA,GAAN,cAAsB,MAAA,CAAO;AAAA;AAAA,EAEjB,CAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQjB,WAAA,CAAY,IAAI,GAAA,EAAK;AACnB,IAAA,KAAA,EAAM;AAGN,IAAA,IAAI,CAAC,OAAO,QAAA,CAAS,CAAC,KAAK,CAAA,GAAI,CAAA,IAAK,KAAK,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIf,uCAAA,CAAsB,CAAA,2CAAA,EAA8C,CAAC,CAAA,CAAA,EAAI,KAAK,CAAC,CAAA;AAAA,IAC3F;AAEA,IAAA,IAAA,CAAK,CAAA,GAAI,CAAA;AAAA,EACX;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,QAAQ,KAAA,EAA8B;AAEpC,IAAA,MAAM,cAAc,KAAA,YAAiBC,4BAAA,GAAa,KAAA,GAAQA,4BAAA,CAAW,WAAW,KAAK,CAAA;AAErF,IAAA,IAAI,WAAA,CAAY,UAAU,QAAA,EAAU;AAClC,MAAA,MAAM,IAAIJ,6BAAW,uCAAuC,CAAA;AAAA,IAC9D;AAIA,IAAA,OAAOmB,yBAAA,CAAY,WAAA,EAAa,IAAA,CAAK,CAAA,EAAG,KAAK,QAAQ,CAAA;AAAA,EACvD;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOS,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,UAAA,EAAa,KAAK,CAAC,CAAA,CAAA,CAAA;AAAA,EAC5B;AAAA;AAAA;AAAA;AAAA,EAKA,IAAI,WAAA,GAAsB;AACxB,IAAA,OAAO,IAAA,CAAK,CAAA;AAAA,EACd;AACF;;;AChCO,IAAM,MAAA,GAAN,cAAqB,MAAA,CAAO;AAAA;AAAA,EAEzB,MAAA;AAAA,EACA,WAAA;AAAA;AAAA,EAGA,IAAA;AAAA,EACA,SAAA;AAAA;AAAA,EAGS,UAAA;AAAA;AAAA,EAGA,WAAA;AAAA;AAAA,EAGA,OAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAYjB,WAAA,CACE,UAAA,EACA,WAAA,EACA,OAAA,GAII,EAAC,EACL;AAEA,IAAA,KAAA,EAAM;AAGN,IAAA,IAAI,cAAc,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,UAAU,CAAA,EAAG;AACpD,MAAA,MAAM,IAAIhB,uCAAA;AAAA,QACR,uCAAA;AAAA,QACA,YAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,eAAe,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,WAAW,CAAA,EAAG;AACtD,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,wCAAA;AAAA,QACA,aAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AAGA,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,WAAA,GAAc,WAAA;AAEnB,IAAA,IAAA,CAAK,OAAA,GAAU,QAAQ,IAAA,IAAQ,IAAA;AAK/B,IAAA,MAAM,MAAA,GAAS,IAAA,CAAK,IAAA,CAAK,CAAA,GAAM,UAAU,CAAA;AAIzC,IAAA,MAAM,YAAA,GAAeiB,uBAAA,CAAM,CAAC,WAAA,EAAa,UAAU,CAAA,EAAG;AAAA,MACpD,KAAA,EAAO,QAAQ,KAAA,IAAS,SAAA;AAAA,MACxB,MAAA,EAAQ,QAAQ,MAAA,IAAU;AAAA,KAC3B,CAAA;AAID,IAAA,MAAM,YAAA,GAAeC,2BAAA,CAAU,YAAA,EAAc,MAAM,CAAA;AACnD,IAAA,IAAA,CAAK,WAAA,GAAcC,4BAAU,YAAY,CAAA;AACzC,IAAA,IAAA,CAAK,MAAA,GAAS,KAAK,WAAA,CAAY,MAAA;AAG/B,IAAA,IAAA,CAAK,iBAAA,CAAkB,QAAA,EAAU,IAAA,CAAK,WAAW,CAAA;AAIjD,IAAA,IAAI,KAAK,OAAA,EAAS;AAEhB,MAAA,MAAM,UAAA,GAAaC,uBAAA,CAAM,CAAC,WAAW,CAAA,EAAG;AAAA,QACtC,KAAA,EAAO,QAAQ,KAAA,IAAS,SAAA;AAAA,QACxB,MAAA,EAAQ,QAAQ,MAAA,IAAU;AAAA,OAC3B,CAAA;AACD,MAAA,IAAA,CAAK,SAAA,GAAYD,4BAAU,UAAU,CAAA;AACrC,MAAA,IAAA,CAAK,IAAA,GAAO,KAAK,SAAA,CAAU,MAAA;AAE3B,MAAA,IAAA,CAAK,iBAAA,CAAkB,MAAA,EAAQ,IAAA,CAAK,SAAS,CAAA;AAAA,IAC/C;AAAA,EACF;AAAA,EAYA,QAAQ,KAAA,EAAiD;AACvD,IAAA,MAAM,WAAA,GAAc,KAAA,YAAiBlB,4BAAA,GAAa,KAAA,CAAM,MAAA,GAAS,KAAA;AAEjE,IAAA,IAAI,WAAA,CAAY,UAAU,QAAA,EAAU;AAClC,MAAA,MAAM,IAAIJ,6BAAW,4CAA4C,CAAA;AAAA,IACnE;AAIA,IAAA,IAAI,WAAA,CAAY,OAAO,CAAA,EAAG;AACxB,MAAA,MAAM,IAAIF,4BAAA,CAAW,CAAA,iDAAA,EAAoD,WAAA,CAAY,IAAI,CAAA,CAAE,CAAA;AAAA,IAC7F;AAIA,IAAA,MAAM,gBAAgB,WAAA,CAAY,KAAA,CAAM,YAAY,KAAA,CAAM,MAAA,GAAS,CAAC,CAAA,IAAK,CAAA;AAGzE,IAAA,IAAI,aAAA,KAAkB,KAAK,UAAA,EAAY;AACrC,MAAA,MAAM,IAAIA,4BAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,IAAA,CAAK,UAAU,CAAA,qBAAA,EAAwB,aAAa,CAAA;AAAA,OAC9E;AAAA,IACF;AAOA,IAAA,MAAM,aAAA,GAAgB,YAAY,IAAA,KAAS,CAAA;AAI3C,IAAA,MAAM,SAAA,GAAY,WAAA,CAAY,IAAA,GAAO,IAAA,CAAK,UAAA;AAG1C,IAAA,MAAM,WAAA,GAAc,aAAA,GAChB,CAAC,IAAA,CAAK,WAAW,CAAA,GACjB,CAAC,GAAG,WAAA,CAAY,MAAM,KAAA,CAAM,CAAA,EAAG,EAAE,CAAA,EAAG,KAAK,WAAW,CAAA;AAExD,IAAA,IAAI,iBAAiBM,4BAAA,EAAY;AAC/B,MAAA,MAAMoB,WAAU,KAAA,CAAM,OAAA,CAAQ,CAAC,SAAA,EAAW,IAAA,CAAK,UAAU,CAAC,CAAA;AAC1D,MAAA,MAAMC,YAAWD,QAAAA,CAAQ,MAAA,CAAO,IAAA,CAAK,WAAA,CAAY,WAAW,CAAA;AAC5D,MAAA,IAAIE,OAAAA,GAASD,SAAAA,CAAS,OAAA,CAAQ,WAAW,CAAA;AACzC,MAAA,IAAI,IAAA,CAAK,OAAA,IAAW,IAAA,CAAK,SAAA,EAAW;AAClC,QAAAC,OAAAA,GAASA,OAAAA,CAAO,GAAA,CAAI,IAAA,CAAK,SAAS,CAAA;AAAA,MACpC;AACA,MAAA,OAAOA,OAAAA;AAAA,IACT;AAEA,IAAA,MAAM,UAAUC,yBAAA,CAAQ,WAAA,EAAa,CAAC,SAAA,EAAW,IAAA,CAAK,UAAU,CAAC,CAAA;AAIjE,IAAA,MAAM,WAAWC,qBAAA,CAAI,OAAA,EAASC,2BAAA,CAAU,IAAA,CAAK,MAAM,CAAC,CAAA;AAIpD,IAAA,MAAM,MAAA,GAASF,yBAAA,CAAQ,QAAA,EAAU,WAAW,CAAA;AAG5C,IAAA,IAAI,IAAA,CAAK,OAAA,IAAW,IAAA,CAAK,IAAA,EAAM;AAG7B,MAAA,OAAOG,qBAAA,CAAI,MAAA,EAAQ,IAAA,CAAK,IAAI,CAAA;AAAA,IAC9B;AAGA,IAAA,OAAO,MAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOS,QAAA,GAAmB;AAC1B,IAAA,MAAM,OAAA,GAAU,IAAA,CAAK,OAAA,GAAU,WAAA,GAAc,YAAA;AAC7C,IAAA,OAAO,sBAAsB,IAAA,CAAK,UAAU,kBAAkB,IAAA,CAAK,WAAW,KAAK,OAAO,CAAA,CAAA,CAAA;AAAA,EAC5F;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,SAAA,GAAoB;AAClB,IAAA,OAAO,IAAA,CAAK,MAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,OAAA,GAA8B;AAC5B,IAAA,OAAO,IAAA,CAAK,IAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA,EAKA,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA,EAKA,IAAI,UAAA,GAAqB;AACvB,IAAA,OAAO,IAAA,CAAK,WAAA;AAAA,EACd;AACF;;;AChRA,SAAS,mBAAmB,CAAA,EAA6B;AACvD,EAAA,IAAIC,8BAAA,CAAa,CAAA,CAAE,KAAA,EAAO,CAAA,CAAE,OAAO,CAAA,EAAG;AACpC,IAAA,OAAO,CAAA;AAAA,EACT;AACA,EAAA,IAAI,CAAA,CAAE,UAAU,QAAA,EAAU;AACxB,IAAA,MAAM,IAAI/B,6BAAW,6CAA6C,CAAA;AAAA,EACpE;AACA,EAAA,MAAM,IAAA,GAAOgC,uCAAA,CAAsB,CAAA,CAAE,KAAK,CAAA;AAC1C,EAAA,MAAM,GAAA,GAAM,IAAI,IAAA,CAAK,CAAA,CAAE,IAAI,CAAA;AAC3B,EAAA,MAAM,cAAA,GAAiB/B,gCAAA,CAAe,CAAA,CAAE,KAAK,CAAA;AAC7C,EAAA,MAAM,OAAO,CAAA,CAAE,IAAA;AAEf,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,MAAM,IAAID,6BAAW,6CAA6C,CAAA;AAAA,EACpE;AAEA,EAAA,IAAI,gBAAgB,aAAA,EAAe;AACjC,IAAA,IAAI,EAAE,eAAe,aAAA,CAAA,EAAgB;AACnC,MAAA,MAAM,IAAIA,6BAAW,+CAA+C,CAAA;AAAA,IACtE;AACA,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,CAAA,CAAE,MAAM,CAAA,EAAA,EAAK;AAC/B,MAAA,MAAM,SAASE,qCAAA,CAAoB,CAAA,EAAG,gBAAgB,CAAA,CAAE,OAAA,EAAS,EAAE,MAAM,CAAA;AACzE,MAAA,GAAA,CAAI,CAAC,CAAA,GAAI+B,kCAAA,CAAiB,IAAA,EAAM,MAAM,CAAA;AAAA,IACxC;AAAA,EACF,CAAA,MAAO;AACL,IAAA,IAAI,eAAe,aAAA,EAAe;AAChC,MAAA,MAAM,IAAIjC,6BAAW,mDAAmD,CAAA;AAAA,IAC1E;AACA,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,CAAA,CAAE,MAAM,CAAA,EAAA,EAAK;AAC/B,MAAA,MAAM,SAASE,qCAAA,CAAoB,CAAA,EAAG,gBAAgB,CAAA,CAAE,OAAA,EAAS,EAAE,MAAM,CAAA;AACzE,MAAA,GAAA,CAAI,CAAC,CAAA,GAAIgC,mCAAA,CAAkB,IAAA,EAAM,MAAM,CAAA;AAAA,IACzC;AAAA,EACF;AAEA,EAAA,OAAOC,yBAAY,cAAA,CAAe;AAAA,IAChC,IAAA,EAAM,GAAA;AAAA,IACN,OAAO,CAAA,CAAE,KAAA;AAAA,IACT,OAAO,CAAA,CAAE,KAAA;AAAA,IACT,QAAQ,CAAA,CAAE;AAAA,GACX,CAAA;AACH;AAwBO,IAAM,WAAA,GAAN,cAA0B,MAAA,CAAO;AAAA,EACrB,WAAA;AAAA,EACA,GAAA;AAAA,EACA,QAAA;AAAA,EACA,MAAA;AAAA,EACA,iBAAA;AAAA,EAET,KAAA;AAAA,EACA,IAAA;AAAA,EACA,WAAA;AAAA,EACA,UAAA;AAAA,EAER,WAAA,CACE,WAAA,EACA,OAAA,GAKI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AACN,IAAA,IACE,CAAC,MAAA,CAAO,QAAA,CAAS,WAAW,CAAA,IAC5B,WAAA,IAAe,CAAA,IACf,IAAA,CAAK,KAAA,CAAM,WAAW,CAAA,KAAM,WAAA,EAC5B;AACA,MAAA,MAAM,IAAIhC,uCAAA;AAAA,QACR,wCAAA;AAAA,QACA,aAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAA,CAAK,WAAA,GAAc,WAAA;AACnB,IAAA,IAAA,CAAK,GAAA,GAAM,QAAQ,GAAA,IAAO,IAAA;AAC1B,IAAA,IAAI,CAAC,OAAO,QAAA,CAAS,IAAA,CAAK,GAAG,CAAA,IAAK,IAAA,CAAK,OAAO,CAAA,EAAG;AAC/C,MAAA,MAAM,IAAIA,uCAAA,CAAsB,+BAAA,EAAiC,KAAA,EAAO,KAAK,GAAG,CAAA;AAAA,IAClF;AACA,IAAA,IAAA,CAAK,QAAA,GAAW,QAAQ,QAAA,IAAY,GAAA;AACpC,IAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,IAAA,CAAK,QAAQ,CAAA,IAAK,IAAA,CAAK,QAAA,GAAW,CAAA,IAAK,IAAA,CAAK,QAAA,GAAW,CAAA,EAAG;AAC7E,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,kCAAA;AAAA,QACA,UAAA;AAAA,QACA,IAAA,CAAK;AAAA,OACP;AAAA,IACF;AACA,IAAA,IAAA,CAAK,MAAA,GAAS,QAAQ,MAAA,IAAU,IAAA;AAChC,IAAA,IAAA,CAAK,iBAAA,GAAoB,QAAQ,iBAAA,IAAqB,IAAA;AAEtD,IAAA,IAAI,KAAK,MAAA,EAAQ;AACf,MAAA,MAAM,KAAA,GAAQiC,sBAAA,CAAK,CAAC,WAAW,CAAC,CAAA;AAChC,MAAA,MAAM,IAAA,GAAOb,uBAAA,CAAM,CAAC,WAAW,CAAC,CAAA;AAChC,MAAA,IAAA,CAAK,KAAA,GAAQD,4BAAU,KAAK,CAAA;AAC5B,MAAA,IAAA,CAAK,IAAA,GAAOA,4BAAU,IAAI,CAAA;AAC1B,MAAA,IAAA,CAAK,iBAAA,CAAkB,QAAA,EAAU,IAAA,CAAK,KAAK,CAAA;AAC3C,MAAA,IAAA,CAAK,iBAAA,CAAkB,MAAA,EAAQ,IAAA,CAAK,IAAI,CAAA;AAAA,IAC1C;AAGA,IAAA,IAAA,CAAK,cAAclB,4BAAA,CAAW,UAAA,CAAWmB,wBAAM,CAAC,WAAW,CAAC,CAAA,EAAG;AAAA,MAC7D,YAAA,EAAc;AAAA,KACf,CAAA;AACD,IAAA,IAAA,CAAK,aAAanB,4BAAA,CAAW,UAAA,CAAWgC,uBAAK,CAAC,WAAW,CAAC,CAAA,EAAG;AAAA,MAC3D,YAAA,EAAc;AAAA,KACf,CAAA;AAED,IAAA,IAAI,KAAK,iBAAA,EAAmB;AAC1B,MAAA,IAAA,CAAK,cAAA,CAAe,cAAA,EAAgB,IAAA,CAAK,WAAA,CAAY,MAAM,CAAA;AAC3D,MAAA,IAAA,CAAK,cAAA,CAAe,aAAA,EAAe,IAAA,CAAK,UAAA,CAAW,MAAM,CAAA;AAAA,IAC3D;AAAA,EACF;AAAA,EAEA,QAAQ,CAAA,EAA0B;AAChC,IAAA,MAAM,QAAQ,CAAA,YAAahC,4BAAA,GAAa,CAAA,GAAIA,4BAAA,CAAW,WAAW,CAAC,CAAA;AAEnE,IAAA,MAAM,aAAa,KAAA,CAAM,KAAA;AACzB,IAAA,IAAI,eAAe,QAAA,EAAU;AAC3B,MAAA,MAAM,IAAIJ,6BAAW,2CAA2C,CAAA;AAAA,IAClE;AAEA,IAAA,IAAI,KAAA,CAAM,IAAA,KAAS,CAAA,IAAK,KAAA,CAAM,SAAS,CAAA,EAAG;AACxC,MAAA,MAAM,IAAIF,4BAAA,CAAW,CAAA,6CAAA,EAAgD,KAAA,CAAM,IAAI,CAAA,CAAE,CAAA;AAAA,IACnF;AAEA,IAAA,MAAM,SAAA,GAAY,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACpC,IAAA,IAAI,SAAA,KAAc,KAAK,WAAA,EAAa;AAClC,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,SAAA,EAAY,KAAK,WAAW,CAAA,eAAA,EAAkB,SAAS,CAAA,CAAE,CAAA;AAAA,IAChF;AAEA,IAAA,MAAM,aAAA,GAAgB,IAAA,CAAK,QAAA,IAAY,CAAC,IAAA,CAAK,iBAAA;AAE7C,IAAA,IAAIuC,KAAAA;AACJ,IAAA,IAAI,MAAA;AAUJ,IAAA,IAAI,aAAA,GAAgB,KAAA;AACpB,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AAEpB,MAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAChC,MAAA,MAAM,MAAA,GAAS,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACjC,MAAA,MAAM,OAAO,KAAA,GAAQ,MAAA;AACrB,MAAA,MAAM,iBAAA,GAAoBC,oCAAA,CAAmB,UAAA,EAAY,aAAa,CAAA;AACtE,MAAA,aAAA,GAAgB,KAAA,CACb,UAAU,CAAC,CAAA,EAAG,GAAG,CAAC,CAAC,CAAA,CACnB,GAAA,CAAIlC,4BAAA,CAAW,MAAA,CAAO,GAAG,EAAE,KAAA,EAAO,mBAAmB,CAAC,EACtD,OAAA,CAAQ,CAAC,IAAA,EAAM,SAAS,CAAC,CAAA;AAAA,IAC9B;AAEA,IAAA,IAAI,aAAA,EAAe;AACjB,MAAA,IAAI,aAAA,CAAc,KAAA,CAAM,CAAC,CAAA,KAAM,CAAA,EAAG;AAChC,QAAA,MAAM,IAAID,uCAAA;AAAA,UACR,yCAAA;AAAA,UACA,OAAA;AAAA,UACA,KAAA,CAAM;AAAA,SACR;AAAA,MACF;AAGA,MAAAkC,KAAAA,GAAO,aAAA,CAAc,IAAA,CAAK,CAAC,CAAA;AAC3B,MAAA,MAAA,GAASE,2BAAA,CAAa,aAAA,EAAe,CAAA,EAAG,CAAC,CAAA;AAEzC,MAAA,IAAI,KAAK,iBAAA,EAAmB;AAC1B,QAAAC,wBAAA,CAAO,MAAM;AAEX,UAAA,MAAM,CAAA,GAAI,aAAA,CAAc,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACpC,UAAA,MAAM,WAAA,GACJ,CAAA,GAAI,CAAA,GAAID,2BAAA,CAAa,aAAA,EAAe,CAAA,EAAG,CAAC,CAAA,GAAIA,2BAAA,CAAa,aAAA,EAAe,CAAA,EAAG,CAAC,CAAA;AAE9E,UAAA,MAAM,IAAI,IAAA,CAAK,QAAA;AACf,UAAA,MAAM,UAAA,GAAa,KAAK,WAAA,CAAY,KAAA;AACpC,UAAA,IAAI,eAAe,QAAA,EAAU;AAC3B,YAAA,MAAM,IAAIvC,6BAAW,8CAA8C,CAAA;AAAA,UACrE;AACA,UAAA,MAAM,SAAA,GAAYI,6BAAW,MAAA,CAAO,CAAA,GAAI,GAAG,EAAE,KAAA,EAAO,YAAY,CAAA;AAChE,UAAA,MAAM,UAAUA,4BAAA,CAAW,MAAA,CAAO,GAAG,EAAE,KAAA,EAAO,YAAY,CAAA;AAG1D,UAAA,MAAM,OAAA,GAAU,IAAA,CAAK,WAAA,CAAY,GAAA,CAAI,SAAS,EAAE,GAAA,CAAIiC,KAAAA,CAAK,GAAA,CAAI,OAAO,CAAC,CAAA;AACrE,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,UAAA,CAAW,GAAA,CAAI,SAAS,EAAE,GAAA,CAAI,WAAA,CAAY,GAAA,CAAI,OAAO,CAAC,CAAA;AAG1E,UAAA,IAAA,CAAK,WAAA,GAAcjC,4BAAA,CAAW,UAAA,CAAW,OAAA,CAAQ,MAAA,EAAQ;AAAA,YACvD,YAAA,EAAc;AAAA,WACf,CAAA;AACD,UAAA,IAAA,CAAK,UAAA,GAAaA,4BAAA,CAAW,UAAA,CAAW,MAAA,CAAO,MAAA,EAAQ;AAAA,YACrD,YAAA,EAAc;AAAA,WACf,CAAA;AAGD,UAAA,IAAA,CAAK,cAAA,CAAe,cAAA,EAAgB,IAAA,CAAK,WAAA,CAAY,MAAM,CAAA;AAC3D,UAAA,IAAA,CAAK,cAAA,CAAe,aAAA,EAAe,IAAA,CAAK,UAAA,CAAW,MAAM,CAAA;AAAA,QAC3D,CAAC,CAAA;AAAA,MACH;AAAA,IACF,CAAA,MAAO;AACL,MAAAiC,QAAO,IAAA,CAAK,WAAA;AACZ,MAAA,MAAA,GAAS,IAAA,CAAK,UAAA;AAAA,IAChB;AAQA,IAAA,IAAI,aAAA,GAAgBA,KAAAA;AACpB,IAAA,IAAI,YAAA,GAAe,MAAA;AAEnB,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,MAAA,aAAA,GAAgBA,MAAK,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAA,EAAW,CAAC,CAAC,CAAA;AAC9C,MAAA,YAAA,GAAe,OAAO,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAA,EAAW,CAAC,CAAC,CAAA;AAAA,IACjD,CAAA,MAAO;AACL,MAAA,aAAA,GAAgBA,KAAAA,CAAK,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAS,CAAC,CAAA;AAC3C,MAAA,YAAA,GAAe,MAAA,CAAO,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAS,CAAC,CAAA;AAAA,IAC9C;AAEA,IAAA,MAAM,SAAA,GAAYjC,6BAAW,MAAA,CAAO,IAAA,CAAK,KAAK,EAAE,KAAA,EAAO,YAAY,CAAA;AACnE,IAAA,MAAM,KAAA,GAAQ,YAAA,CAAa,GAAA,CAAI,SAAS,EAAE,IAAA,EAAK;AAC/C,IAAA,IAAI,MAAM,KAAA,CAAM,GAAA,CAAI,aAAa,CAAA,CAAE,IAAI,KAAK,CAAA;AAE5C,IAAA,IAAI,IAAA,CAAK,MAAA,IAAU,IAAA,CAAK,KAAA,IAAS,KAAK,IAAA,EAAM;AAC1C,MAAA,IAAI,SAAS,IAAA,CAAK,KAAA;AAClB,MAAA,IAAI,QAAQ,IAAA,CAAK,IAAA;AACjB,MAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,QAAA,MAAA,GAAS,KAAK,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAA,EAAW,CAAC,CAAC,CAAA;AAC7C,QAAA,KAAA,GAAQ,KAAK,IAAA,CAAK,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAA,EAAW,CAAC,CAAC,CAAA;AAAA,MAC7C,CAAA,MAAO;AACL,QAAA,MAAA,GAAS,KAAK,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAS,CAAC,CAAA;AAC1C,QAAA,KAAA,GAAQ,KAAK,IAAA,CAAK,OAAA,CAAQ,CAAC,CAAA,EAAG,SAAS,CAAC,CAAA;AAAA,MAC1C;AACA,MAAA,GAAA,GAAM,GAAA,CAAI,GAAA,CAAI,MAAM,CAAA,CAAE,IAAI,KAAK,CAAA;AAAA,IACjC;AAEA,IAAA,OAAO,GAAA;AAAA,EACT;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,YAAA,EAAe,IAAA,CAAK,WAAW,CAAA,MAAA,EAAS,IAAA,CAAK,GAAG,CAAA,WAAA,EAAc,IAAA,CAAK,QAAQ,CAAA,SAAA,EAAY,IAAA,CAAK,MAAM,CAAA,CAAA,CAAA;AAAA,EAC3G;AACF;AAuBO,IAAM,SAAA,GAAN,cAAwB,MAAA,CAAO;AAAA,EACnB,eAAA;AAAA,EACA,GAAA;AAAA,EACA,iBAAA;AAAA,EAET,KAAA;AAAA,EACA,IAAA;AAAA,EAER,WAAA,CACE,eAAA,EACA,OAAA,GAGI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AACN,IAAA,IAAA,CAAK,eAAA,GACH,OAAO,eAAA,KAAoB,QAAA,GAAW,CAAC,eAAe,CAAA,GAAI,KAAA,CAAM,IAAA,CAAK,eAAe,CAAA;AAEtF,IAAA,IAAI,IAAA,CAAK,eAAA,CAAgB,MAAA,KAAW,CAAA,EAAG;AACrC,MAAA,MAAM,IAAID,uCAAA;AAAA,QACR,qDAAA;AAAA,QACA,iBAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AAEA,IAAA,KAAA,MAAW,GAAA,IAAO,KAAK,eAAA,EAAiB;AACtC,MAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,GAAG,CAAA,IAAK,GAAA,IAAO,CAAA,IAAK,IAAA,CAAK,KAAA,CAAM,GAAG,CAAA,KAAM,GAAA,EAAK;AAChE,QAAA,MAAM,IAAIA,uCAAA;AAAA,UACR,6DAAA;AAAA,UACA,iBAAA;AAAA,UACA;AAAA,SACF;AAAA,MACF;AAAA,IACF;AAEA,IAAA,IAAA,CAAK,GAAA,GAAM,QAAQ,GAAA,IAAO,IAAA;AAC1B,IAAA,IAAI,CAAC,OAAO,QAAA,CAAS,IAAA,CAAK,GAAG,CAAA,IAAK,IAAA,CAAK,OAAO,CAAA,EAAG;AAC/C,MAAA,MAAM,IAAIA,uCAAA,CAAsB,+BAAA,EAAiC,KAAA,EAAO,KAAK,GAAG,CAAA;AAAA,IAClF;AAEA,IAAA,IAAA,CAAK,iBAAA,GAAoB,QAAQ,iBAAA,IAAqB,IAAA;AAEtD,IAAA,IAAI,KAAK,iBAAA,EAAmB;AAC1B,MAAA,IAAA,CAAK,KAAA,GAAQmB,2BAAA,CAAUc,sBAAA,CAAK,IAAA,CAAK,eAAe,CAAC,CAAA;AACjD,MAAA,IAAA,CAAK,IAAA,GAAOd,2BAAA,CAAUC,uBAAA,CAAM,IAAA,CAAK,eAAe,CAAC,CAAA;AACjD,MAAA,IAAA,CAAK,iBAAA,CAAkB,QAAA,EAAU,IAAA,CAAK,KAAK,CAAA;AAC3C,MAAA,IAAA,CAAK,iBAAA,CAAkB,MAAA,EAAQ,IAAA,CAAK,IAAI,CAAA;AAAA,IAC1C;AAAA,EACF;AAAA,EAEA,QAAQ,CAAA,EAA0B;AAChC,IAAA,MAAM,QAAQ,CAAA,YAAanB,4BAAA,GAAa,CAAA,GAAIA,4BAAA,CAAW,WAAW,CAAC,CAAA;AAEnE,IAAA,MAAM,aAAa,KAAA,CAAM,KAAA;AACzB,IAAA,IAAI,eAAe,QAAA,EAAU;AAC3B,MAAA,MAAM,IAAIJ,6BAAW,yCAAyC,CAAA;AAAA,IAChE;AAEA,IAAA,IAAI,YAAA,GAAe,KAAA;AACnB,IAAA,IAAI,CAAC+B,+BAAa,KAAA,CAAM,MAAA,CAAO,OAAO,KAAA,CAAM,MAAA,CAAO,OAAO,CAAA,EAAG;AAC3D,MAAA,MAAM,UAAA,GAAa,kBAAA,CAAmB,KAAA,CAAM,MAAM,CAAA;AAClD,MAAA,YAAA,GAAe3B,4BAAA,CAAW,WAAW,UAAA,EAAY;AAAA,QAC/C,cAAc,KAAA,CAAM;AAAA,OACrB,CAAA;AAAA,IACH;AAGA,IAAA,MAAM,aAAa,YAAA,CAAa,KAAA;AAChC,IAAA,MAAM,YAAY,IAAA,CAAK,eAAA;AACvB,IAAA,IAAI,SAAA,CAAU,MAAA,GAAS,UAAA,CAAW,MAAA,EAAQ;AACxC,MAAA,MAAM,IAAIN,4BAAA,CAAW,CAAA,YAAA,EAAe,UAAU,CAAA,+BAAA,EAAkC,SAAS,CAAA,CAAE,CAAA;AAAA,IAC7F;AAGA,IAAA,MAAM,WAAA,GAAc,UAAA,CAAW,MAAA,GAAS,SAAA,CAAU,MAAA;AAClD,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,SAAA,CAAU,QAAQ,CAAA,EAAA,EAAK;AACzC,MAAA,IAAI,WAAW,WAAA,GAAc,CAAC,CAAA,KAAM,SAAA,CAAU,CAAC,CAAA,EAAG;AAChD,QAAA,MAAM,IAAIA,4BAAA;AAAA,UACR,CAAA,YAAA,EAAe,UAAU,CAAA,mCAAA,EAAsC,SAAS,CAAA;AAAA,SAC1E;AAAA,MACF;AAAA,IACF;AAOA,IAAA,MAAM,SAAA,GAAY,UAAA,CAAW,KAAA,CAAM,CAAA,EAAG,WAAW,CAAA;AACjD,IAAA,MAAM,QAAA,GAAW,UAAU,MAAA,CAAO,CAAC,GAAG,CAAA,KAAM,CAAA,GAAI,GAAG,CAAC,CAAA;AAEpD,IAAA,MAAM,cAAA,GAAiB,CAAC,GAAG,SAAA,EAAW,QAAQ,CAAA;AAC9C,IAAA,MAAM,aAAA,GAAgB,YAAA,CAAa,OAAA,CAAQ,cAAc,CAAA;AAGzD,IAAA,MAAMuC,KAAAA,GAAO,aAAA,CAAc,IAAA,CAAK,EAAA,EAAI,IAAI,CAAA;AACxC,IAAA,MAAM,MAAA,GAASE,2BAAA,CAAa,aAAA,EAAe,EAAA,EAAI,CAAC,CAAA;AAUhD,IAAA,MAAM,WAAA,GAAc,MAAA,CAAO,OAAA,CAAQF,KAAAA,CAAK,KAAK,CAAA;AAG7C,IAAA,MAAM,SAAA,GAAYjC,6BAAW,MAAA,CAAO,IAAA,CAAK,KAAK,EAAE,KAAA,EAAO,YAAY,CAAA;AACnE,IAAA,MAAM,KAAA,GAAQ,WAAA,CAAY,GAAA,CAAI,SAAS,EAAE,IAAA,EAAK;AAC9C,IAAA,MAAM,qBAAqB,aAAA,CAAc,GAAA,CAAIiC,KAAI,CAAA,CAAE,IAAI,KAAK,CAAA;AAG5D,IAAA,IAAI,GAAA,GAAM,kBAAA,CAAmB,OAAA,CAAQ,UAAU,CAAA;AAG/C,IAAA,IAAI,IAAA,CAAK,iBAAA,IAAqB,IAAA,CAAK,KAAA,IAAS,KAAK,IAAA,EAAM;AAIrD,MAAA,GAAA,GAAM,IAAI,GAAA,CAAI,IAAA,CAAK,KAAK,CAAA,CAAE,GAAA,CAAI,KAAK,IAAI,CAAA;AAAA,IACzC;AAEA,IAAA,OAAO,GAAA;AAAA,EACT;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,UAAA,EAAa,KAAK,eAAe,CAAA,MAAA,EAAS,KAAK,GAAG,CAAA,qBAAA,EAAwB,KAAK,iBAAiB,CAAA,CAAA,CAAA;AAAA,EACzG;AACF;;;ACpZO,IAAM,kBAAA,GAAN,cAAiC,MAAA,CAAO;AAAA;AAAA,EAE5B,QAAA;AAAA;AAAA,EAGA,QAAA;AAAA;AAAA,EAGA,OAAA;AAAA;AAAA,EAGA,KAAA;AAAA;AAAA,EAGA,OAAA;AAAA;AAAA,EAGA,OAAA;AAAA;AAAA,EAGT,EAAA;AAAA,EACA,EAAA;AAAA;AAAA,EAGA,EAAA;AAAA,EACA,EAAA;AAAA;AAAA,EAGA,EAAA;AAAA,EACA,EAAA;AAAA;AAAA,EAGA,EAAA;AAAA,EACA,EAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAWR,WAAA,CACE,QAAA,EACA,QAAA,EACA,OAAA,GAGI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AAEN,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,QAAQ,CAAA,IAAK,YAAY,CAAA,EAAG;AAChD,MAAA,MAAM,IAAIlC,uCAAA,CAAsB,qCAAA,EAAuC,UAAA,EAAY,QAAQ,CAAA;AAAA,IAC7F;AACA,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,QAAQ,CAAA,IAAK,YAAY,CAAA,EAAG;AAChD,MAAA,MAAM,IAAIA,uCAAA,CAAsB,qCAAA,EAAuC,UAAA,EAAY,QAAQ,CAAA;AAAA,IAC7F;AACA,IAAA,IAAI,QAAA,GAAW,aAAa,CAAA,EAAG;AAC7B,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,CAAA,UAAA,EAAa,QAAQ,CAAA,iCAAA,EAAoC,QAAQ,CAAA,CAAA,CAAA;AAAA,QACjE,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AAEA,IAAA,MAAMgB,QAAAA,GAAU,QAAQ,OAAA,IAAW,CAAA;AACnC,IAAA,IAAI,CAAC,OAAO,QAAA,CAASA,QAAO,KAAKA,QAAAA,GAAU,CAAA,IAAKA,YAAW,CAAA,EAAG;AAC5D,MAAA,MAAM,IAAIhB,uCAAA,CAAsB,2BAAA,EAA6B,SAAA,EAAWgB,QAAO,CAAA;AAAA,IACjF;AAEA,IAAA,IAAA,CAAK,QAAA,GAAW,QAAA;AAChB,IAAA,IAAA,CAAK,QAAA,GAAW,QAAA;AAChB,IAAA,IAAA,CAAK,UAAU,QAAA,GAAW,QAAA;AAC1B,IAAA,IAAA,CAAK,KAAA,GAAQ,IAAA,CAAK,IAAA,CAAK,IAAA,CAAK,OAAO,CAAA;AACnC,IAAA,IAAA,CAAK,OAAA,GAAU,QAAQ,IAAA,IAAQ,IAAA;AAC/B,IAAA,IAAA,CAAK,OAAA,GAAUA,QAAAA;AAGf,IAAA,MAAM,MAAA,GAAS,IAAA,CAAK,IAAA,CAAK,CAAA,IAAO,WAAW,QAAA,CAAS,CAAA;AAIpD,IAAA,IAAA,CAAK,EAAA,GAAKG,2BAAA,CAAUD,2BAAA,CAAUD,uBAAA,CAAM,CAAC,UAAU,QAAQ,CAAC,CAAA,EAAG,MAAM,CAAC,CAAA;AAClE,IAAA,IAAA,CAAK,EAAA,GAAKE,2BAAA,CAAUD,2BAAA,CAAUD,uBAAA,CAAM,CAAC,UAAU,QAAQ,CAAC,CAAA,EAAG,MAAM,CAAC,CAAA;AAClE,IAAA,IAAA,CAAK,EAAA,GAAKE,2BAAA,CAAUD,2BAAA,CAAUD,uBAAA,CAAM,CAAC,UAAU,QAAQ,CAAC,CAAA,EAAG,MAAM,CAAC,CAAA;AAClE,IAAA,IAAA,CAAK,EAAA,GAAKE,2BAAA,CAAUD,2BAAA,CAAUD,uBAAA,CAAM,CAAC,UAAU,QAAQ,CAAC,CAAA,EAAG,MAAM,CAAC,CAAA;AAElE,IAAA,IAAA,CAAK,iBAAA,CAAkB,kBAAA,EAAoB,IAAA,CAAK,EAAE,CAAA;AAClD,IAAA,IAAA,CAAK,iBAAA,CAAkB,kBAAA,EAAoB,IAAA,CAAK,EAAE,CAAA;AAClD,IAAA,IAAA,CAAK,iBAAA,CAAkB,kBAAA,EAAoB,IAAA,CAAK,EAAE,CAAA;AAClD,IAAA,IAAA,CAAK,iBAAA,CAAkB,iBAAA,EAAmB,IAAA,CAAK,EAAE,CAAA;AAEjD,IAAA,IAAI,KAAK,OAAA,EAAS;AAChB,MAAA,IAAA,CAAK,KAAKE,2BAAA,CAAUC,uBAAA,CAAM,CAAC,QAAQ,CAAC,CAAC,CAAA;AACrC,MAAA,IAAA,CAAK,KAAKD,2BAAA,CAAUC,uBAAA,CAAM,CAAC,QAAQ,CAAC,CAAC,CAAA;AACrC,MAAA,IAAA,CAAK,KAAKD,2BAAA,CAAUC,uBAAA,CAAM,CAAC,QAAQ,CAAC,CAAC,CAAA;AACrC,MAAA,IAAA,CAAK,KAAKD,2BAAA,CAAUC,uBAAA,CAAM,CAAC,QAAQ,CAAC,CAAC,CAAA;AAErC,MAAA,IAAA,CAAK,iBAAA,CAAkB,gBAAA,EAAkB,IAAA,CAAK,EAAE,CAAA;AAChD,MAAA,IAAA,CAAK,iBAAA,CAAkB,gBAAA,EAAkB,IAAA,CAAK,EAAE,CAAA;AAChD,MAAA,IAAA,CAAK,iBAAA,CAAkB,gBAAA,EAAkB,IAAA,CAAK,EAAE,CAAA;AAChD,MAAA,IAAA,CAAK,iBAAA,CAAkB,eAAA,EAAiB,IAAA,CAAK,EAAE,CAAA;AAAA,IACjD;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUA,WAAW,MAAA,EAAiC;AAC1C,IAAA,IAAI,MAAA,CAAO,MAAA,GAAS,CAAA,IAAK,MAAA,CAAO,SAAS,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIpB,uCAAA;AAAA,QACR,yDAAA;AAAA,QACA,QAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAAA,IACF;AAEA,IAAA,MAAM,UAAA,GAAa,OAAO,CAAC,CAAA;AAC3B,IAAA,IAAI,eAAe,MAAA,EAAW;AAC5B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,0BAAA,EAA4B,OAAA,EAAS,UAAU,CAAA;AAAA,IACjF;AAGA,IAAA,MAAM,QAAQ,UAAA,YAAsBC,4BAAA,GAAa,UAAA,GAAaA,4BAAA,CAAW,WAAW,UAAU,CAAA;AAE9F,IAAA,MAAM,QAAA,GAAW,MAAA,CAAO,CAAC,CAAA,IAAK,UAAA;AAC9B,IAAA,MAAM,MAAM,QAAA,YAAoBA,4BAAA,GAAa,QAAA,GAAWA,4BAAA,CAAW,WAAW,QAAQ,CAAA;AAEtF,IAAA,MAAM,UAAA,GAAa,MAAA,CAAO,CAAC,CAAA,IAAK,UAAA;AAChC,IAAA,MAAM,QAAQ,UAAA,YAAsBA,4BAAA,GAAa,UAAA,GAAaA,4BAAA,CAAW,WAAW,UAAU,CAAA;AAE9F,IAAA,IAAI,MAAM,KAAA,KAAU,QAAA,EAAU,MAAM,IAAIJ,6BAAW,kCAAkC,CAAA;AACrF,IAAA,IAAI,MAAM,IAAA,KAAS,GAAA,CAAI,QAAQ,KAAA,CAAM,IAAA,KAAS,MAAM,IAAA,EAAM;AACxD,MAAA,MAAM,IAAIF,6BAAW,2CAA2C,CAAA;AAAA,IAClE;AACA,IAAA,IAAI,KAAA,CAAM,IAAA,KAAS,CAAA,IAAK,KAAA,CAAM,SAAS,CAAA,EAAG;AACxC,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,iCAAA,EAAoC,KAAA,CAAM,IAAI,CAAA,CAAE,CAAA;AAAA,IACvE;AACA,IAAA,IAAI,GAAA,CAAI,IAAA,KAAS,CAAA,IAAK,GAAA,CAAI,SAAS,CAAA,EAAG;AACpC,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,+BAAA,EAAkC,GAAA,CAAI,IAAI,CAAA,CAAE,CAAA;AAAA,IACnE;AACA,IAAA,IAAI,KAAA,CAAM,IAAA,KAAS,CAAA,IAAK,KAAA,CAAM,SAAS,CAAA,EAAG;AACxC,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,iCAAA,EAAoC,KAAA,CAAM,IAAI,CAAA,CAAE,CAAA;AAAA,IACvE;AAKA,IAAA,IAAI,CAAA,GAAI,KAAA;AACR,IAAA,IAAI,CAAA,GAAI,GAAA;AACR,IAAA,IAAI,CAAA,GAAI,KAAA;AAER,IAAA,IAAI,EAAE,IAAA,KAAS,CAAA,MAAO,CAAA,CAAE,OAAA,CAAQ,CAAC,CAAA,EAAG,CAAA,CAAE,KAAA,CAAM,CAAC,KAAK,CAAA,EAAG,CAAA,CAAE,MAAM,CAAC,CAAA,IAAK,CAAC,CAAC,CAAA;AACrE,IAAA,IAAI,EAAE,IAAA,KAAS,CAAA,MAAO,CAAA,CAAE,OAAA,CAAQ,CAAC,CAAA,EAAG,CAAA,CAAE,KAAA,CAAM,CAAC,KAAK,CAAA,EAAG,CAAA,CAAE,MAAM,CAAC,CAAA,IAAK,CAAC,CAAC,CAAA;AACrE,IAAA,IAAI,EAAE,IAAA,KAAS,CAAA,MAAO,CAAA,CAAE,OAAA,CAAQ,CAAC,CAAA,EAAG,CAAA,CAAE,KAAA,CAAM,CAAC,KAAK,CAAA,EAAG,CAAA,CAAE,MAAM,CAAC,CAAA,IAAK,CAAC,CAAC,CAAA;AAErE,IAAA,MAAM,SAAA,GAAY,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAChC,IAAA,MAAM,OAAA,GAAU,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,OAAA,GAAU,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,OAAA,GAAU,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,QAAA,GAAW,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAE/B,IAAA,IAAI,QAAA,KAAa,KAAK,QAAA,EAAU;AAC9B,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,kCAAA,EAAqC,KAAK,QAAQ,CAAA,MAAA,EAAS,QAAQ,CAAA,CAAE,CAAA;AAAA,IAC5F;AACA,IAAA,IAAI,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,KAAM,KAAK,QAAA,EAAU;AAChC,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,gCAAA,EAAmC,IAAA,CAAK,QAAQ,SAAS,CAAA,CAAE,KAAA,CAAM,CAAC,CAAC,CAAA,CAAE,CAAA;AAAA,IAC5F;AACA,IAAA,IAAI,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,KAAM,KAAK,QAAA,EAAU;AAChC,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,kCAAA,EAAqC,IAAA,CAAK,QAAQ,SAAS,CAAA,CAAE,KAAA,CAAM,CAAC,CAAC,CAAA,CAAE,CAAA;AAAA,IAC9F;AACA,IAAA,IAAI,CAAA,CAAE,MAAM,CAAC,CAAA,KAAM,aAAa,CAAA,CAAE,KAAA,CAAM,CAAC,CAAA,KAAM,SAAA,EAAW;AACxD,MAAA,MAAM,IAAIA,4BAAA;AAAA,QACR,CAAA,2BAAA,EAA8B,SAAS,CAAA,MAAA,EAAS,CAAA,CAAE,KAAA,CAAM,CAAC,CAAC,CAAA,QAAA,EAAW,CAAA,CAAE,KAAA,CAAM,CAAC,CAAC,CAAA;AAAA,OACjF;AAAA,IACF;AACA,IAAA,IAAI,YAAY,OAAA,EAAS;AACvB,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,wCAAA,EAA2C,OAAO,CAAA,QAAA,EAAW,OAAO,CAAA,CAAE,CAAA;AAAA,IAC7F;AAKA,IAAA,IAAI,IAAI,CAAA,CAAE,MAAA,CAAO,IAAA,CAAK,EAAA,CAAG,WAAW,CAAA;AACpC,IAAA,IAAI,KAAK,EAAA,EAAI,CAAA,GAAI,CAAA,CAAE,GAAA,CAAI,KAAK,EAAE,CAAA;AAE9B,IAAA,IAAI,IAAI,CAAA,CAAE,MAAA,CAAO,IAAA,CAAK,EAAA,CAAG,WAAW,CAAA;AACpC,IAAA,IAAI,KAAK,EAAA,EAAI,CAAA,GAAI,CAAA,CAAE,GAAA,CAAI,KAAK,EAAE,CAAA;AAE9B,IAAA,IAAI,IAAI,CAAA,CAAE,MAAA,CAAO,IAAA,CAAK,EAAA,CAAG,WAAW,CAAA;AACpC,IAAA,IAAI,KAAK,EAAA,EAAI,CAAA,GAAI,CAAA,CAAE,GAAA,CAAI,KAAK,EAAE,CAAA;AAI9B,IAAA,MAAM,IAAI,IAAA,CAAK,QAAA;AACf,IAAA,MAAM,IAAI,IAAA,CAAK,OAAA;AAEf,IAAA,CAAA,GAAI,CAAA,CAAE,OAAA,CAAQ,CAAC,SAAA,EAAW,SAAS,CAAA,EAAG,CAAC,CAAC,CAAA,CAAE,UAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAA;AAChE,IAAA,CAAA,GAAI,CAAA,CAAE,OAAA,CAAQ,CAAC,SAAA,EAAW,SAAS,CAAA,EAAG,CAAC,CAAC,CAAA,CAAE,UAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAA;AAChE,IAAA,CAAA,GAAI,CAAA,CAAE,OAAA,CAAQ,CAAC,SAAA,EAAW,SAAS,CAAA,EAAG,CAAC,CAAC,CAAA,CAAE,UAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAA;AAKhE,IAAA,IAAI,MAAA,GAAS,CAAA,CAAE,MAAA,CAAO,CAAA,CAAE,SAAA,CAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAC,CAAA;AAC/C,IAAA,MAAA,GAAS,OAAO,GAAA,CAAIM,4BAAA,CAAW,MAAA,CAAO,IAAA,CAAK,KAAK,CAAC,CAAA;AAGjD,IAAA,IAAI,IAAA,GAAOS,0BAAAA,CAAY,MAAA,EAAQ,EAAE,CAAA;AAGjC,IAAA,IAAA,GAAOM,yBAAA,CAAY,IAAA,EAAM,IAAA,CAAK,OAAA,EAAS,KAAK,QAAQ,CAAA;AAKpD,IAAA,MAAM,OAAA,GAAU,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAI7B,IAAA,MAAM,YAAA,GAAemB,oCAAA,CAAmB,OAAA,CAAQ,KAAA,EAAO,oBAAoB,CAAA;AAC3E,IAAA,MAAM,eAAA,GAAkB,OAAA,CACrB,SAAA,CAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAA,CACtB,GAAA,CAAIlC,4BAAA,CAAW,MAAA,CAAO,GAAG,EAAE,KAAA,EAAO,YAAA,EAAc,CAAC,CAAA,CACjD,OAAA,CAAQ,CAAC,SAAA,EAAW,OAAA,EAAS,IAAA,CAAK,QAAQ,CAAC,CAAA;AAG9C,IAAA,IAAI,SAAS,eAAA,CAAgB,MAAA,CAAO,IAAA,CAAK,EAAA,CAAG,WAAW,CAAA;AACvD,IAAA,IAAI,KAAK,EAAA,EAAI,MAAA,GAAS,MAAA,CAAO,GAAA,CAAI,KAAK,EAAE,CAAA;AAGxC,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,MAAA,MAAA,GAAS,OAAO,OAAA,CAAQ,CAAC,OAAA,EAAS,IAAA,CAAK,QAAQ,CAAC,CAAA;AAAA,IAClD;AAEA,IAAA,OAAO,MAAA;AAAA,EACT;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,6BAAA,EAAgC,IAAA,CAAK,QAAQ,CAAA,YAAA,EAAe,KAAK,QAAQ,CAAA,CAAA,CAAA;AAAA,EAClF;AACF;AAuBO,IAAM,uBAAA,GAAN,cAAsC,MAAA,CAAO;AAAA,EACjC,MAAA;AAAA,EACA,KAAA;AAAA,EACA,GAAA;AAAA,EAEA,QAAA;AAAA,EACA,OAAA;AAAA,EACA,OAAA;AAAA,EACA,KAAA;AAAA,EACA,KAAA;AAAA,EAEA,OAAA;AAAA;AAAA;AAAA,EAGA,QAAA;AAAA,EACA,QAAA;AAAA,EACA,QAAA;AAAA,EAEjB,YACE,MAAA,EACA,KAAA,EACA,GAAA,EACA,OAAA,GAGI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AAEN,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,MAAM,CAAA,IAAK,UAAU,CAAA,EAAG;AAC5C,MAAA,MAAM,IAAID,uCAAA,CAAsB,mCAAA,EAAqC,QAAA,EAAU,MAAM,CAAA;AAAA,IACvF;AACA,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,IAAK,SAAS,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIA,uCAAA,CAAsB,kCAAA,EAAoC,OAAA,EAAS,KAAK,CAAA;AAAA,IACpF;AACA,IAAA,IAAI,MAAA,GAAS,UAAU,CAAA,EAAG;AACxB,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,CAAA,QAAA,EAAW,MAAM,CAAA,8BAAA,EAAiC,KAAK,CAAA,CAAA,CAAA;AAAA,QACvD,QAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,GAAG,CAAA,IAAK,OAAO,CAAA,EAAG;AACtC,MAAA,MAAM,IAAIA,uCAAA,CAAsB,gCAAA,EAAkC,KAAA,EAAO,GAAG,CAAA;AAAA,IAC9E;AAEA,IAAA,MAAMgB,QAAAA,GAAU,QAAQ,OAAA,IAAW,GAAA;AACnC,IAAA,MAAM,GAAA,GAAM,QAAQ,GAAA,IAAO,IAAA;AAE3B,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AACd,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AACb,IAAA,IAAA,CAAK,GAAA,GAAM,GAAA;AACX,IAAA,IAAA,CAAK,OAAA,GAAUA,QAAAA;AAEf,IAAA,IAAA,CAAK,QAAA,GAAW,IAAI,kBAAA,CAAmB,MAAA,EAAQ,OAAO,EAAE,OAAA,EAAAA,UAAS,CAAA;AACjE,IAAA,IAAA,CAAK,OAAA,GAAU,IAAI,MAAA,CAAO,MAAA,EAAQ,GAAG,CAAA;AACrC,IAAA,IAAA,CAAK,OAAA,GAAU,IAAI,MAAA,CAAO,GAAA,EAAK,MAAM,CAAA;AACrC,IAAA,IAAA,CAAK,QAAQ,IAAI,SAAA,CAAU,MAAA,EAAQ,EAAE,KAAK,CAAA;AAC1C,IAAA,IAAA,CAAK,QAAQ,IAAI,SAAA,CAAU,MAAA,EAAQ,EAAE,KAAK,CAAA;AAC1C,IAAA,IAAA,CAAK,QAAA,GAAW,IAAI,OAAA,CAAQA,QAAO,CAAA;AACnC,IAAA,IAAA,CAAK,QAAA,GAAW,IAAI,OAAA,CAAQA,QAAO,CAAA;AACnC,IAAA,IAAA,CAAK,QAAA,GAAW,IAAI,OAAA,CAAQA,QAAO,CAAA;AAEnC,IAAA,IAAA,CAAK,cAAA,CAAe,WAAA,EAAa,IAAA,CAAK,QAAQ,CAAA;AAC9C,IAAA,IAAA,CAAK,cAAA,CAAe,SAAA,EAAW,IAAA,CAAK,OAAO,CAAA;AAC3C,IAAA,IAAA,CAAK,cAAA,CAAe,SAAA,EAAW,IAAA,CAAK,OAAO,CAAA;AAC3C,IAAA,IAAA,CAAK,cAAA,CAAe,OAAA,EAAS,IAAA,CAAK,KAAK,CAAA;AACvC,IAAA,IAAA,CAAK,cAAA,CAAe,OAAA,EAAS,IAAA,CAAK,KAAK,CAAA;AACvC,IAAA,IAAA,CAAK,cAAA,CAAe,UAAA,EAAY,IAAA,CAAK,QAAQ,CAAA;AAC7C,IAAA,IAAA,CAAK,cAAA,CAAe,UAAA,EAAY,IAAA,CAAK,QAAQ,CAAA;AAC7C,IAAA,IAAA,CAAK,cAAA,CAAe,UAAA,EAAY,IAAA,CAAK,QAAQ,CAAA;AAAA,EAC/C;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,QAAQ,GAAA,EAA4B;AAClC,IAAA,MAAM,QAAQ,GAAA,YAAef,4BAAA,GAAa,GAAA,GAAMA,4BAAA,CAAW,WAAW,GAAG,CAAA;AAEzE,IAAA,IAAI,KAAA,CAAM,UAAU,QAAA,EAAU;AAC5B,MAAA,MAAM,IAAIJ,6BAAW,uDAAuD,CAAA;AAAA,IAC9E;AAIA,IAAA,IAAI,OAAO,IAAA,CAAK,QAAA,CAAS,OAAA,CAAQ,KAAA,EAAO,OAAO,KAAK,CAAA;AAGpD,IAAA,IAAA,GAAO,IAAA,CAAK,QAAA,CAAS,OAAA,CAAQ,IAAI,CAAA;AACjC,IAAA,IAAI,GAAA,GAAM,KAAA,CAAM,GAAA,CAAI,IAAI,CAAA;AAGxB,IAAA,GAAA,GAAM,IAAA,CAAK,KAAA,CAAM,OAAA,CAAQ,GAAG,CAAA;AAK5B,IAAA,IAAI,GAAA,GAAM,IAAA,CAAK,OAAA,CAAQ,OAAA,CAAQ,GAAG,CAAA;AAClC,IAAA,GAAA,GAAM,IAAI,IAAA,EAAK;AACf,IAAA,GAAA,GAAM,IAAA,CAAK,QAAA,CAAS,OAAA,CAAQ,GAAG,CAAA;AAC/B,IAAA,GAAA,GAAM,IAAA,CAAK,OAAA,CAAQ,OAAA,CAAQ,GAAG,CAAA;AAG9B,IAAA,GAAA,GAAM,IAAA,CAAK,QAAA,CAAS,OAAA,CAAQ,GAAG,CAAA;AAC/B,IAAA,GAAA,GAAM,GAAA,CAAI,IAAI,GAAG,CAAA;AAGjB,IAAA,GAAA,GAAM,IAAA,CAAK,KAAA,CAAM,OAAA,CAAQ,GAAG,CAAA;AAE5B,IAAA,OAAO,GAAA;AAAA,EACT;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,gCAAA,EAAmC,IAAA,CAAK,MAAM,CAAA,QAAA,EAAW,IAAA,CAAK,KAAK,CAAA,kBAAA,EAAqB,IAAA,CAAK,GAAG,CAAA,UAAA,EAAa,IAAA,CAAK,OAAO,CAAA,CAAA,CAAA;AAAA,EAClI;AACF;;;AC5aA,SAAS,aAAA,CACP,IAAA,EACA,KAAA,EACA,SAAA,EACA,WAAA,EACkB;AAClB,EAAA,MAAM,MAAM,OAAO,KAAA,KAAU,WAAW,CAAC,KAAA,EAAO,KAAK,CAAA,GAAI,KAAA;AACzD,EAAA,MAAM,KAAA,GAAQ,IAAI,CAAC,CAAA;AACnB,EAAA,MAAM,MAAA,GAAS,IAAI,CAAC,CAAA;AACpB,EAAA,IACE,GAAA,CAAI,MAAA,KAAW,CAAA,IACf,KAAA,KAAU,MAAA,IACV,WAAW,MAAA,IACX,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,IACvB,CAAC,MAAA,CAAO,SAAA,CAAU,MAAM,CAAA,KACvB,SAAA,GAAY,KAAA,GAAQ,CAAA,IAAK,MAAA,GAAS,CAAA,GAAI,KAAA,IAAS,CAAA,IAAK,MAAA,IAAU,CAAA,CAAA,EAC/D;AACA,IAAA,MAAM,IAAIG,wCAAsB,CAAA,EAAG,IAAI,YAAY,WAAW,CAAA,CAAA,EAAI,MAAM,KAAK,CAAA;AAAA,EAC/E;AACA,EAAA,OAAO,CAAC,OAAO,MAAM,CAAA;AACvB;AAgBO,IAAM,MAAA,GAAN,cAAqB,MAAA,CAAO;AAAA,EAChB,UAAA;AAAA,EACA,WAAA;AAAA,EACA,UAAA;AAAA,EACA,MAAA;AAAA,EACA,OAAA;AAAA,EACA,IAAA;AAAA,EAET,OAAA;AAAA,EACA,KAAA;AAAA,EAER,YACE,UAAA,EACA,WAAA,EACA,UAAA,EACA,OAAA,GAII,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AAGN,IAAA,IAAI,cAAc,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,UAAU,CAAA,EAAG;AACpD,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,uCAAA;AAAA,QACA,YAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,eAAe,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,WAAW,CAAA,EAAG;AACtD,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,wCAAA;AAAA,QACA,aAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,cAAc,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,UAAU,CAAA,EAAG;AACpD,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,uCAAA;AAAA,QACA,YAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AAEA,IAAA,MAAM,MAAA,GAAS,QAAQ,MAAA,IAAU,CAAA;AACjC,IAAA,IAAI,UAAU,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,MAAM,CAAA,EAAG;AAC5C,MAAA,MAAM,IAAIA,uCAAA,CAAsB,mCAAA,EAAqC,QAAA,EAAU,MAAM,CAAA;AAAA,IACvF;AAEA,IAAA,MAAM,OAAA,GAAU,QAAQ,OAAA,IAAW,CAAA;AACnC,IAAA,IAAI,UAAU,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,OAAO,CAAA,EAAG;AAC7C,MAAA,MAAM,IAAIA,uCAAA,CAAsB,wCAAA,EAA0C,SAAA,EAAW,OAAO,CAAA;AAAA,IAC9F;AAEA,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,WAAA,GAAc,WAAA;AACnB,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AACd,IAAA,IAAA,CAAK,OAAA,GAAU,OAAA;AACf,IAAA,IAAA,CAAK,IAAA,GAAO,QAAQ,IAAA,IAAQ,IAAA;AAE5B,IAAA,IAAA,CAAK,oBAAA,EAAqB;AAAA,EAC5B;AAAA,EAEQ,oBAAA,GAA6B;AACnC,IAAA,MAAM,IAAI,CAAA,GAAI,IAAA,CAAK,KAAK,IAAA,CAAK,UAAA,GAAa,KAAK,UAAU,CAAA;AACzD,IAAA,MAAM,MAAA,GAASiB,wBAAM,CAAC,IAAA,CAAK,aAAa,IAAA,CAAK,UAAA,EAAY,IAAA,CAAK,UAAU,CAAC,CAAA;AACzE,IAAA,IAAA,CAAK,OAAA,GAAUE,2BAAA,CAAUD,2BAAA,CAAU,MAAA,EAAQ,CAAC,CAAC,CAAA;AAC7C,IAAA,IAAA,CAAK,iBAAA,CAAkB,QAAA,EAAU,IAAA,CAAK,OAAO,CAAA;AAE7C,IAAA,IAAI,KAAK,IAAA,EAAM;AACb,MAAA,MAAM,QAAA,GAAWD,uBAAA,CAAM,CAAC,IAAA,CAAK,WAAW,CAAC,CAAA;AACzC,MAAA,IAAA,CAAK,KAAA,GAAQE,2BAAA,CAAUD,2BAAA,CAAU,QAAA,EAAU,CAAC,CAAC,CAAA;AAC7C,MAAA,IAAA,CAAK,iBAAA,CAAkB,MAAA,EAAQ,IAAA,CAAK,KAAK,CAAA;AAAA,IAC3C;AAAA,EACF;AAAA,EAEA,QAAQ,CAAA,EAA0B;AAEhC,IAAA,MAAM,QAAQ,CAAA,YAAajB,4BAAA,GAAa,CAAA,GAAIA,4BAAA,CAAW,WAAW,CAAC,CAAA;AAGnE,IAAA,IAAI,KAAA,CAAM,UAAU,QAAA,EAAU;AAC5B,MAAA,MAAM,IAAIJ,6BAAW,kCAAkC,CAAA;AAAA,IACzD;AAGA,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,MAAA,MAAM,IAAIF,4BAAA,CAAW,CAAA,uDAAA,EAA0D,KAAA,CAAM,IAAI,CAAA,CAAA,CAAG,CAAA;AAAA,IAC9F;AAEA,IAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAChC,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAE9B,IAAA,IAAI,GAAA,KAAQ,KAAK,UAAA,EAAY;AAC3B,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,SAAA,EAAY,KAAK,UAAU,CAAA,qBAAA,EAAwB,GAAG,CAAA,CAAE,CAAA;AAAA,IAC/E;AAEA,IAAA,MAAM,SAAS,IAAA,CAAK,OAAA;AACpB,IAAA,IAAI,CAAC,MAAA,EAAQ,MAAM,IAAI2C,iCAAe,wBAAwB,CAAA;AAI9D,IAAA,MAAM,OAAA,GAAU,MAAM,OAAA,CAAQ,CAAC,OAAO,GAAA,EAAK,CAAA,EAAG,GAAG,CAAC,CAAA;AAIlD,IAAA,MAAM,UAAA,GAA+B,CAAC,CAAA,EAAG,IAAA,CAAK,UAAU,CAAA;AACxD,IAAA,MAAM,MAAA,GAA2B,CAAC,CAAA,EAAG,IAAA,CAAK,MAAM,CAAA;AAChD,IAAA,MAAM,OAAA,GAA4B,CAAC,CAAA,EAAG,IAAA,CAAK,OAAO,CAAA;AAGlD,IAAA,MAAM,IAAA,GAAOC,yBAAA,CAAW,OAAA,EAAS,UAAA,EAAY,QAAQ,OAAO,CAAA;AAK5D,IAAA,MAAM,UAAA,GAAa,MAAA,CAAO,OAAA,CAAQ,CAAC,IAAA,CAAK,aAAa,IAAA,CAAK,UAAA,GAAa,IAAA,CAAK,UAAU,CAAC,CAAA;AAGvF,IAAA,MAAM,GAAA,GAAM,IAAA,CAAK,MAAA,CAAO,UAAA,CAAW,WAAW,CAAA;AAG9C,IAAA,MAAM,gBAAgB,GAAA,CAAI,SAAA,CAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAA;AAE7C,IAAA,IAAI,IAAA,CAAK,IAAA,IAAQ,IAAA,CAAK,KAAA,EAAO;AAE3B,MAAA,MAAM,YAAA,GAAe,KAAK,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,EAAG,IAAA,CAAK,WAAA,EAAa,CAAC,CAAC,CAAA;AAChE,MAAA,OAAO,aAAA,CAAc,IAAI,YAAY,CAAA;AAAA,IACvC;AAEA,IAAA,OAAO,aAAA;AAAA,EACT;AAAA,EAEA,IAAI,MAAA,GAAqB;AACvB,IAAA,IAAI,CAAC,KAAK,OAAA,EAAS;AACjB,MAAA,MAAM,IAAID,iCAAe,wBAAwB,CAAA;AAAA,IACnD;AACA,IAAA,OAAO,IAAA,CAAK,OAAA;AAAA,EACd;AACF;AAgBO,IAAM,MAAA,GAAN,cAAqB,MAAA,CAAO;AAAA,EAChB,UAAA;AAAA,EACA,WAAA;AAAA,EACA,UAAA;AAAA,EACA,MAAA;AAAA,EACA,OAAA;AAAA,EACA,OAAA;AAAA,EAET,OAAA;AAAA,EACA,KAAA;AAAA,EAER,YACE,UAAA,EACA,WAAA,EACA,UAAA,EACA,OAAA,GAII,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AACN,IAAA,IAAI,cAAc,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,UAAU,CAAA,EAAG;AACpD,MAAA,MAAM,IAAItC,uCAAA;AAAA,QACR,uCAAA;AAAA,QACA,YAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,eAAe,CAAA,IAAK,CAAC,MAAA,CAAO,SAAA,CAAU,WAAW,CAAA,EAAG;AACtD,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,wCAAA;AAAA,QACA,aAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,MAAM,SAAA,GAAY,aAAA;AAAA,MAChB,YAAA;AAAA,MACA,UAAA;AAAA,MACA,KAAA;AAAA,MACA;AAAA,KACF;AAEA,IAAA,MAAM,MAAA,GAAS,QAAQ,MAAA,IAAU,CAAA;AACjC,IAAA,MAAM,SAAA,GAAY,aAAA;AAAA,MAChB,QAAA;AAAA,MACA,MAAA;AAAA,MACA,KAAA;AAAA,MACA;AAAA,KACF;AAEA,IAAA,MAAM,OAAA,GAAU,QAAQ,OAAA,IAAW,CAAA;AACnC,IAAA,MAAM,UAAA,GAAa,aAAA;AAAA,MACjB,SAAA;AAAA,MACA,OAAA;AAAA,MACA,IAAA;AAAA,MACA;AAAA,KACF;AAEA,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,WAAA,GAAc,WAAA;AACnB,IAAA,IAAA,CAAK,UAAA,GAAa,SAAA;AAClB,IAAA,IAAA,CAAK,MAAA,GAAS,SAAA;AACd,IAAA,IAAA,CAAK,OAAA,GAAU,UAAA;AAEf,IAAA,IAAA,CAAK,OAAA,GAAU,QAAQ,IAAA,IAAQ,IAAA;AAE/B,IAAA,IAAA,CAAK,oBAAA,EAAqB;AAAA,EAC5B;AAAA,EAEQ,oBAAA,GAA6B;AACnC,IAAA,MAAM,EAAA,GAAK,IAAA,CAAK,UAAA,CAAW,CAAC,CAAA,IAAK,CAAA;AACjC,IAAA,MAAM,EAAA,GAAK,IAAA,CAAK,UAAA,CAAW,CAAC,CAAA,IAAK,CAAA;AACjC,IAAA,MAAM,IAAI,CAAA,GAAI,IAAA,CAAK,KAAK,IAAA,CAAK,UAAA,GAAa,KAAK,EAAE,CAAA;AACjD,IAAA,MAAM,MAAA,GAASiB,wBAAM,CAAC,IAAA,CAAK,aAAa,IAAA,CAAK,UAAA,EAAY,EAAA,EAAI,EAAE,CAAC,CAAA;AAChE,IAAA,IAAA,CAAK,OAAA,GAAUE,2BAAA,CAAUD,2BAAA,CAAU,MAAA,EAAQ,CAAC,CAAC,CAAA;AAC7C,IAAA,IAAA,CAAK,iBAAA,CAAkB,QAAA,EAAU,IAAA,CAAK,OAAO,CAAA;AAE7C,IAAA,IAAI,KAAK,OAAA,EAAS;AAChB,MAAA,MAAM,QAAA,GAAWD,uBAAA,CAAM,CAAC,IAAA,CAAK,WAAW,CAAC,CAAA;AACzC,MAAA,IAAA,CAAK,KAAA,GAAQE,2BAAA,CAAUD,2BAAA,CAAU,QAAA,EAAU,CAAC,CAAC,CAAA;AAC7C,MAAA,IAAA,CAAK,iBAAA,CAAkB,MAAA,EAAQ,IAAA,CAAK,KAAK,CAAA;AAAA,IAC3C;AAAA,EACF;AAAA,EAEA,QAAQ,CAAA,EAA0B;AAChC,IAAA,MAAM,QAAQ,CAAA,YAAajB,4BAAA,GAAa,CAAA,GAAIA,4BAAA,CAAW,WAAW,CAAC,CAAA;AAEnE,IAAA,IAAI,KAAA,CAAM,UAAU,QAAA,EAAU;AAC5B,MAAA,MAAM,IAAIJ,6BAAW,kCAAkC,CAAA;AAAA,IACzD;AAEA,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,MAAA,MAAM,IAAIF,4BAAA;AAAA,QACR,CAAA,8DAAA,EAAiE,MAAM,IAAI,CAAA,CAAA;AAAA,OAC7E;AAAA,IACF;AAEA,IAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAChC,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAE9B,IAAA,IAAI,GAAA,KAAQ,KAAK,UAAA,EAAY;AAC3B,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,SAAA,EAAY,KAAK,UAAU,CAAA,qBAAA,EAAwB,GAAG,CAAA,CAAE,CAAA;AAAA,IAC/E;AAEA,IAAA,MAAM,SAAS,IAAA,CAAK,OAAA;AACpB,IAAA,IAAI,CAAC,MAAA,EAAQ,MAAM,IAAI2C,iCAAe,wBAAwB,CAAA;AAE9D,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,UAAA;AACtB,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,MAAA;AACtB,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,OAAA;AAGtB,IAAA,MAAM,IAAA,GAAOC,yBAAA,CAAW,KAAA,EAAO,CAAC,IAAI,EAAE,CAAA,EAAG,CAAC,EAAA,EAAI,EAAE,CAAA,EAAG,CAAC,EAAA,EAAI,EAAE,CAAC,CAAA;AAG3D,IAAA,MAAM,IAAA,GAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,IAAI,EAAA,GAAK,EAAA,IAAM,EAAE,CAAA,GAAI,CAAA;AACpD,IAAA,MAAM,IAAA,GAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,IAAI,EAAA,GAAK,EAAA,IAAM,EAAE,CAAA,GAAI,CAAA;AAGpD,IAAA,MAAM,UAAA,GAAa,MAAA,CAAO,OAAA,CAAQ,CAAC,IAAA,CAAK,aAAa,IAAA,CAAK,UAAA,GAAa,EAAA,GAAK,EAAE,CAAC,CAAA;AAG/E,IAAA,MAAM,GAAA,GAAM,IAAA,CAAK,MAAA,CAAO,UAAA,CAAW,WAAW,CAAA;AAG9C,IAAA,MAAM,gBAAgB,GAAA,CAAI,SAAA,CAAU,CAAC,CAAA,EAAG,CAAA,EAAG,CAAC,CAAC,CAAA;AAC7C,IAAA,MAAM,WAAA,GAAc,cAAc,OAAA,CAAQ,CAAC,OAAO,IAAA,CAAK,WAAA,EAAa,IAAA,EAAM,IAAI,CAAC,CAAA;AAE/E,IAAA,IAAI,IAAA,CAAK,OAAA,IAAW,IAAA,CAAK,KAAA,EAAO;AAE9B,MAAA,MAAM,YAAA,GAAe,IAAA,CAAK,KAAA,CAAM,OAAA,CAAQ,CAAC,GAAG,IAAA,CAAK,WAAA,EAAa,CAAA,EAAG,CAAC,CAAC,CAAA;AACnE,MAAA,OAAO,WAAA,CAAY,IAAI,YAAY,CAAA;AAAA,IACrC;AAEA,IAAA,OAAO,WAAA;AAAA,EACT;AAAA,EAEA,IAAI,MAAA,GAAqB;AACvB,IAAA,IAAI,CAAC,KAAK,OAAA,EAAS;AACjB,MAAA,MAAM,IAAID,iCAAe,wBAAwB,CAAA;AAAA,IACnD;AACA,IAAA,OAAO,IAAA,CAAK,OAAA;AAAA,EACd;AACF;AAgBO,IAAM,SAAA,GAAN,cAAwB,MAAA,CAAO;AAAA,EACnB,eAAA;AAAA,EACA,MAAA;AAAA,EACA,OAAA;AAAA,EAEjB,WAAA,CACE,UAAA,EACA,OAAA,GAGI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AAEN,IAAA,MAAM,SAAA,GAAY,aAAA;AAAA,MAChB,YAAA;AAAA,MACA,UAAA;AAAA,MACA,KAAA;AAAA,MACA;AAAA,KACF;AACA,IAAA,IAAA,CAAK,eAAA,GAAkB,SAAA;AAEvB,IAAA,MAAM,SAAA,GAAY,aAAA;AAAA,MAChB,QAAA;AAAA,MACA,QAAQ,MAAA,IAAU,UAAA;AAAA,MAClB,KAAA;AAAA,MACA;AAAA,KACF;AACA,IAAA,IAAA,CAAK,MAAA,GAAS,SAAA;AAEd,IAAA,MAAM,UAAA,GAAa,aAAA;AAAA,MACjB,SAAA;AAAA,MACA,QAAQ,OAAA,IAAW,CAAA;AAAA,MACnB,IAAA;AAAA,MACA;AAAA,KACF;AACA,IAAA,IAAA,CAAK,OAAA,GAAU,UAAA;AAAA,EACjB;AAAA,EAEA,QAAQ,CAAA,EAA0B;AAChC,IAAA,MAAM,QAAQ,CAAA,YAAarC,4BAAA,GAAa,CAAA,GAAIA,4BAAA,CAAW,WAAW,CAAC,CAAA;AAEnE,IAAA,IAAI,KAAA,CAAM,UAAU,QAAA,EAAU;AAC5B,MAAA,MAAM,IAAIJ,6BAAW,kCAAkC,CAAA;AAAA,IACzD;AAEA,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,MAAA,MAAM,IAAIF,4BAAA;AAAA,QACR,CAAA,iEAAA,EAAoE,MAAM,IAAI,CAAA,CAAA;AAAA,OAChF;AAAA,IACF;AAEA,IAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAChC,IAAA,MAAM,QAAA,GAAW,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACnC,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAE9B,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,eAAA;AACtB,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,MAAA;AACtB,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,OAAA;AAMtB,IAAA,MAAM,aAAA,GAAgB,MAAM,OAAA,CAAQ,CAAC,QAAQ,QAAA,EAAU,CAAA,EAAG,GAAA,EAAK,GAAG,CAAC,CAAA;AAGnE,IAAA,MAAM,IAAA,GAAO4C,yBAAA,CAAW,aAAA,EAAe,CAAC,IAAI,EAAE,CAAA,EAAG,CAAC,EAAA,EAAI,EAAE,CAAA,EAAG,CAAC,EAAA,EAAI,EAAE,CAAC,CAAA;AAInE,IAAA,MAAM,OAAA,GAAU,IAAA,CAAK,GAAA,CAAI,CAAC,CAAA;AAG1B,IAAA,MAAM,IAAA,GAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,IAAI,EAAA,GAAK,EAAA,IAAM,EAAE,CAAA,GAAI,CAAA;AACpD,IAAA,MAAM,IAAA,GAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,IAAI,EAAA,GAAK,EAAA,IAAM,EAAE,CAAA,GAAI,CAAA;AAGpD,IAAA,OAAO,QAAQ,OAAA,CAAQ,CAAC,OAAO,QAAA,EAAU,IAAA,EAAM,IAAI,CAAC,CAAA;AAAA,EACtD;AACF;AAgBO,IAAM,SAAA,GAAN,cAAwB,MAAA,CAAO;AAAA,EACnB,eAAA;AAAA,EACA,MAAA;AAAA,EACA,OAAA;AAAA,EAEjB,WAAA,CACE,UAAA,EACA,OAAA,GAGI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AAEN,IAAA,MAAM,SAAA,GAAY,aAAA;AAAA,MAChB,YAAA;AAAA,MACA,UAAA;AAAA,MACA,KAAA;AAAA,MACA;AAAA,KACF;AACA,IAAA,IAAA,CAAK,eAAA,GAAkB,SAAA;AAEvB,IAAA,MAAM,SAAA,GAAY,aAAA;AAAA,MAChB,QAAA;AAAA,MACA,QAAQ,MAAA,IAAU,UAAA;AAAA,MAClB,KAAA;AAAA,MACA;AAAA,KACF;AACA,IAAA,IAAA,CAAK,MAAA,GAAS,SAAA;AAEd,IAAA,MAAM,UAAA,GAAa,aAAA;AAAA,MACjB,SAAA;AAAA,MACA,QAAQ,OAAA,IAAW,CAAA;AAAA,MACnB,IAAA;AAAA,MACA;AAAA,KACF;AACA,IAAA,IAAA,CAAK,OAAA,GAAU,UAAA;AAAA,EACjB;AAAA,EAEA,QAAQ,CAAA,EAA0B;AAChC,IAAA,MAAM,QAAQ,CAAA,YAAatC,4BAAA,GAAa,CAAA,GAAIA,4BAAA,CAAW,WAAW,CAAC,CAAA;AAEnE,IAAA,IAAI,KAAA,CAAM,UAAU,QAAA,EAAU;AAC5B,MAAA,MAAM,IAAIJ,6BAAW,kCAAkC,CAAA;AAAA,IACzD;AAEA,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,MAAA,MAAM,IAAIF,4BAAA;AAAA,QACR,CAAA,iEAAA,EAAoE,MAAM,IAAI,CAAA,CAAA;AAAA,OAChF;AAAA,IACF;AAEA,IAAA,MAAM,KAAA,GAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAChC,IAAA,MAAM,QAAA,GAAW,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACnC,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAC9B,IAAA,MAAM,GAAA,GAAM,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAE9B,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,eAAA;AACtB,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,MAAA;AACtB,IAAA,MAAM,CAAC,EAAA,EAAI,EAAE,CAAA,GAAI,IAAA,CAAK,OAAA;AAGtB,IAAA,MAAM,aAAA,GAAgB,MAAM,OAAA,CAAQ,CAAC,QAAQ,QAAA,EAAU,CAAA,EAAG,GAAA,EAAK,GAAG,CAAC,CAAA;AAGnE,IAAA,MAAM,IAAA,GAAO4C,yBAAA,CAAW,aAAA,EAAe,CAAC,IAAI,EAAE,CAAA,EAAG,CAAC,EAAA,EAAI,EAAE,CAAA,EAAG,CAAC,EAAA,EAAI,EAAE,CAAC,CAAA;AAInE,IAAA,MAAM,QAAA,GAAW,IAAA,CAAK,IAAA,CAAK,CAAC,CAAA;AAG5B,IAAA,MAAM,IAAA,GAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,IAAI,EAAA,GAAK,EAAA,IAAM,EAAE,CAAA,GAAI,CAAA;AACpD,IAAA,MAAM,IAAA,GAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,IAAI,EAAA,GAAK,EAAA,IAAM,EAAE,CAAA,GAAI,CAAA;AAGpD,IAAA,OAAO,SAAS,OAAA,CAAQ,CAAC,OAAO,QAAA,EAAU,IAAA,EAAM,IAAI,CAAC,CAAA;AAAA,EACvD;AACF;;;ACtgBA,SAAS,iBAAA,CACP,GACA,OAAA,EACmD;AACnD,EAAA,IAAI,CAAA,CAAE,UAAU,QAAA,EAAU;AACxB,IAAA,MAAM,IAAI1C,4BAAA,CAAW,CAAA,EAAG,OAAO,CAAA,8BAAA,CAAgC,CAAA;AAAA,EACjE;AACA,EAAA,IAAI,CAAA,CAAE,KAAA,KAAU,SAAA,IAAa,CAAA,CAAE,UAAU,SAAA,EAAW;AAClD,IAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,EAAG,OAAO,CAAA,iCAAA,CAAmC,CAAA;AAAA,EACpE;AACF;AAEA,SAAS,WAAA,CAAY,GAAW,MAAA,EAAwB;AACtD,EAAA,MAAM,OAAO,CAAA,CAAE,IAAA;AACf,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,MAAM,IAAIA,6BAAW,kCAAkC,CAAA;AAAA,EACzD;AACA,EAAA,OAAO2C,oCAAA,CAAmB,MAAM,MAAM,CAAA;AACxC;AAEA,SAAS,iBAAA,CACP,MACA,KAAA,EAC6B;AAC7B,EAAA,OAAO,KAAA,KAAU,YAAY,IAAI,YAAA,CAAa,IAAI,CAAA,GAAI,IAAI,aAAa,IAAI,CAAA;AAC7E;AAEA,SAAS,mBAAA,CAAoB,MAAc,KAAA,EAAqB;AAC9D,EAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,IAAK,SAAS,CAAA,EAAG;AAC1C,IAAA,MAAM,IAAIxC,uCAAA,CAAsB,CAAA,EAAG,IAAI,CAAA,2BAAA,CAAA,EAA+B,MAAM,KAAK,CAAA;AAAA,EACnF;AACF;AAYA,SAAS,UAAA,CAAW,OAAe,UAAA,EAAkC;AACnE,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,MAAM,MAAA,GAAS,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACjC,IAAA,MAAM,QAAA,GAAW,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACnC,IAAA,OAAO;AAAA,MACL,KAAA,EAAO,CAAA;AAAA,MACP,MAAA;AAAA,MACA,QAAA;AAAA,MACA,WAAA,EAAa,IAAA;AAAA,MACb,WAAA,EAAa,CAAA;AAAA,MACb,SAAA,EAAW,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AAAA,MAC/B,UAAA,EAAY,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK;AAAA,KAClC;AAAA,EACF;AAEA,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,MAAM,IAAIL,4BAAA,CAAW,CAAA,iDAAA,EAAoD,KAAA,CAAM,IAAI,CAAA,CAAE,CAAA;AAAA,EACvF;AAEA,EAAA,IAAI,UAAA,EAAY;AACd,IAAA,OAAO;AAAA,MACL,KAAA,EAAO,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,MACzB,MAAA,EAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,MAC1B,QAAA,EAAU,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,MAC5B,WAAA,EAAa,KAAA;AAAA,MACb,WAAA,EAAa,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AAAA,MACjC,SAAA,EAAW,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AAAA,MAC/B,UAAA,EAAY,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK;AAAA,KAClC;AAAA,EACF;AAEA,EAAA,OAAO;AAAA,IACL,KAAA,EAAO,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,IACzB,MAAA,EAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,IAC1B,QAAA,EAAU,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,IAC5B,WAAA,EAAa,KAAA;AAAA,IACb,WAAA,EAAa,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AAAA,IACjC,SAAA,EAAW,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AAAA,IAC/B,UAAA,EAAY,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK;AAAA,GAClC;AACF;AAEA,SAAS,WAAA,CACP,YACA,WAAA,EACA,KAAA,EACA,QACA,UAAA,EACA,CAAA,EACA,GACA,CAAA,EACQ;AACR,EAAA,IAAI,WAAA,EAAa;AACf,IAAA,OAAO,IAAI,UAAA,GAAa,CAAA;AAAA,EAC1B;AACA,EAAA,IAAI,UAAA,EAAY;AACd,IAAA,OAAO,CAAA,IAAK,MAAA,GAAS,UAAA,CAAA,GAAc,CAAA,GAAI,UAAA,GAAa,CAAA;AAAA,EACtD;AACA,EAAA,OAAO,CAAA,IAAK,KAAA,GAAQ,UAAA,CAAA,GAAc,CAAA,GAAI,UAAA,GAAa,CAAA;AACrD;AAEA,SAAS,aAAA,CAAc,KAAgB,KAAA,EAAuB;AAC5D,EAAA,IAAI,eAAeM,4BAAA,EAAY;AAC7B,IAAA,OAAO,GAAA,CAAI,MAAA;AAAA,EACb;AACA,EAAA,OAAO,GAAA;AACT;AAEA,SAAS,WACP,KAAA,EACA,SAAA,EACA,KAAA,EACA,UAAA,EACA,aACA,IAAA,EACgB;AAChB,EAAA,MAAM,MAAA,GAAS,IAAI,KAAA,CAAoB,SAAS,CAAA;AAChD,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,SAAA,EAAW,CAAA,EAAA,EAAK;AAClC,IAAA,MAAA,CAAO,CAAC,CAAA,GAAI,IAAI,YAAA,CAAa,QAAQ,UAAU,CAAA;AAAA,EACjD;AAEA,EAAA,IAAI,CAAC,KAAA,EAAO;AACV,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,iBAAA,CAAkB,OAAO,IAAI,CAAA;AAE7B,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,IAAI,CAAC,WAAA,EAAa;AAChB,MAAA,MAAM,IAAIN,4BAAA,CAAW,CAAA,SAAA,EAAY,IAAI,CAAA,oCAAA,CAAsC,CAAA;AAAA,IAC7E;AACA,IAAA,IAAA,CAAK,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA,MAAO,SAAA,IAAA,CAAc,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA,MAAO,UAAA,EAAY;AAC/E,MAAA,MAAM,IAAIA,4BAAA;AAAA,QACR,CAAA,SAAA,EAAY,IAAI,CAAA,QAAA,EAAW,SAAS,CAAA,EAAA,EAAK,UAAU,CAAA,QAAA,EAAW,KAAA,CAAM,KAAA,CAAM,IAAA,CAAK,IAAI,CAAC,CAAA,CAAA;AAAA,OACtF;AAAA,IACF;AAEA,IAAA,MAAM8C,QAAAA,GAAU,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACpC,IAAA,MAAMC,QAAAA,GAAU,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACpC,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,SAAA,EAAW,CAAA,EAAA,EAAK;AAClC,MAAA,MAAM,UAAA,GAAa,OAAO,CAAC,CAAA;AAC3B,MAAA,IAAI,CAAC,UAAA,EAAY;AACf,QAAA,MAAM,IAAI/C,4BAAA,CAAW,CAAA,wBAAA,EAA2B,IAAI,CAAA,YAAA,CAAc,CAAA;AAAA,MACpE;AACA,MAAA,MAAM,IAAA,GAAO,KAAA,CAAM,MAAA,GAAS,CAAA,GAAI8C,QAAAA;AAChC,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,UAAA,EAAY,CAAA,EAAA,EAAK;AACnC,QAAA,UAAA,CAAW,CAAC,CAAA,GAAI,WAAA,CAAY,KAAA,EAAO,IAAA,GAAO,IAAIC,QAAO,CAAA;AAAA,MACvD;AAAA,IACF;AAEA,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,MAAM,IAAI/C,4BAAA,CAAW,CAAA,SAAA,EAAY,IAAI,CAAA,kCAAA,EAAqC,KAAA,CAAM,IAAI,CAAA,CAAE,CAAA;AAAA,EACxF;AAEA,EAAA,MAAM,aAAA,GAAgB,cAAc,CAAA,GAAI,KAAA;AACxC,EAAA,IAAA,CACG,MAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA,MAAO,cACzB,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,OAAO,aAAA,IAAA,CACzB,KAAA,CAAM,MAAM,CAAC,CAAA,IAAK,OAAO,UAAA,EAC1B;AACA,IAAA,MAAM,QAAA,GAAW,WAAA,GAAc,CAAC,SAAA,EAAW,CAAA,EAAG,UAAU,CAAA,GAAI,CAAC,SAAA,EAAW,KAAA,EAAO,UAAU,CAAA;AACzF,IAAA,MAAM,IAAIA,4BAAA;AAAA,MACR,CAAA,SAAA,EAAY,IAAI,CAAA,QAAA,EAAW,QAAA,CAAS,IAAA,CAAK,IAAI,CAAC,CAAA,QAAA,EAAW,KAAA,CAAM,KAAA,CAAM,IAAA,CAAK,IAAI,CAAC,CAAA,CAAA;AAAA,KACjF;AAAA,EACF;AAEA,EAAA,MAAM,OAAA,GAAU,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACpC,EAAA,MAAM,OAAA,GAAU,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACpC,EAAA,MAAM,OAAA,GAAU,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AAEpC,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,SAAA,EAAW,CAAA,EAAA,EAAK;AAClC,IAAA,MAAM,UAAA,GAAa,OAAO,CAAC,CAAA;AAC3B,IAAA,IAAI,CAAC,UAAA,EAAY;AACf,MAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,wBAAA,EAA2B,IAAI,CAAA,YAAA,CAAc,CAAA;AAAA,IACpE;AACA,IAAA,MAAM,SAAA,GAAY,KAAA,CAAM,MAAA,GAAS,CAAA,GAAI,OAAA;AACrC,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,KAAA,EAAO,CAAA,EAAA,EAAK;AAC9B,MAAA,MAAM,SAAA,GAAY,YAAY,CAAA,GAAI,OAAA;AAClC,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,UAAA,EAAY,CAAA,EAAA,EAAK;AACnC,QAAA,UAAA,CAAW,CAAA,GAAI,aAAa,CAAC,CAAA,GAAI,YAAY,KAAA,EAAO,SAAA,GAAY,IAAI,OAAO,CAAA;AAAA,MAC7E;AAAA,IACF;AAAA,EACF;AAEA,EAAA,OAAO,MAAA;AACT;AAEA,SAAS,UACP,KAAA,EACA,SAAA,EACA,OACA,UAAA,EACA,KAAA,EACA,QACA,WAAA,EACQ;AACR,EAAA,MAAM,IAAA,GAAO,WAAA,GAAc,SAAA,GAAY,UAAA,GAAa,YAAY,KAAA,GAAQ,UAAA;AACxE,EAAA,MAAM,IAAA,GAAO,iBAAA,CAAkB,IAAA,EAAM,KAAK,CAAA;AAE1C,EAAA,IAAI,WAAA,EAAa;AACf,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,SAAA,EAAW,CAAA,EAAA,EAAK;AAClC,MAAA,MAAM,KAAA,GAAQ,MAAM,CAAC,CAAA;AACrB,MAAA,IAAI,CAAC,KAAA,EAAO;AACV,QAAA,MAAM,IAAIA,6BAAW,4CAA4C,CAAA;AAAA,MACnE;AACA,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,UAAA,EAAY,CAAA,EAAA,EAAK;AACnC,QAAA,IAAA,CAAK,IAAI,UAAA,GAAa,CAAC,CAAA,GAAI,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAAA,MACzC;AAAA,IACF;AACA,IAAA,OAAOqC,yBAAO,cAAA,CAAe;AAAA,MAC3B,IAAA;AAAA,MACA,KAAA,EAAO,CAAC,SAAA,EAAW,UAAU,CAAA;AAAA,MAC7B,KAAA;AAAA,MACA;AAAA,KACD,CAAA;AAAA,EACH;AAEA,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,SAAA,EAAW,CAAA,EAAA,EAAK;AAClC,IAAA,MAAM,KAAA,GAAQ,MAAM,CAAC,CAAA;AACrB,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAIrC,6BAAW,4CAA4C,CAAA;AAAA,IACnE;AACA,IAAA,MAAM,WAAA,GAAc,IAAI,KAAA,GAAQ,UAAA;AAChC,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,KAAA,EAAO,CAAA,EAAA,EAAK;AAC9B,MAAA,MAAM,WAAA,GAAc,cAAc,CAAA,GAAI,UAAA;AACtC,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,UAAA,EAAY,CAAA,EAAA,EAAK;AACnC,QAAA,IAAA,CAAK,cAAc,CAAC,CAAA,GAAI,MAAM,CAAA,GAAI,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA;AAAA,MACvD;AAAA,IACF;AAAA,EACF;AAEA,EAAA,OAAOqC,yBAAO,cAAA,CAAe;AAAA,IAC3B,IAAA;AAAA,IACA,KAAA,EAAO,CAAC,SAAA,EAAW,KAAA,EAAO,UAAU,CAAA;AAAA,IACpC,KAAA;AAAA,IACA;AAAA,GACD,CAAA;AACH;AAqBO,IAAM,GAAA,GAAN,cAAkB,MAAA,CAAO;AAAA,EACb,SAAA;AAAA,EACA,UAAA;AAAA,EACA,SAAA;AAAA,EACA,YAAA;AAAA,EACA,IAAA;AAAA,EACA,UAAA;AAAA,EAET,SAAA;AAAA,EACA,SAAA;AAAA,EACA,MAAA;AAAA,EACA,MAAA;AAAA,EAER,WAAA,CACE,SAAA,EACA,UAAA,EACA,OAAA,GAKI,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AACN,IAAA,mBAAA,CAAoB,aAAa,SAAS,CAAA;AAC1C,IAAA,mBAAA,CAAoB,cAAc,UAAU,CAAA;AAC5C,IAAA,MAAM,SAAA,GAAY,QAAQ,SAAA,IAAa,CAAA;AACvC,IAAA,mBAAA,CAAoB,aAAa,SAAS,CAAA;AAE1C,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,YAAA,GAAe,QAAQ,YAAA,IAAgB,MAAA;AAC5C,IAAA,IAAA,CAAK,IAAA,GAAO,QAAQ,IAAA,IAAQ,IAAA;AAC5B,IAAA,IAAA,CAAK,UAAA,GAAa,QAAQ,UAAA,IAAc,IAAA;AAExC,IAAA,MAAM,IAAA,GAAO,CAAA,GAAM,IAAA,CAAK,IAAA,CAAK,UAAU,CAAA;AAEvC,IAAA,IAAA,CAAK,YAAY,EAAC;AAClB,IAAA,IAAA,CAAK,YAAY,EAAC;AAClB,IAAA,IAAA,CAAK,SAAS,EAAC;AACf,IAAA,IAAA,CAAK,SAAS,EAAC;AAEf,IAAA,KAAA,IAAS,KAAA,GAAQ,CAAA,EAAG,KAAA,GAAQ,IAAA,CAAK,WAAW,KAAA,EAAA,EAAS;AACnD,MAAA,MAAM,QAAA,GAAW,KAAA,KAAU,CAAA,GAAI,SAAA,GAAY,UAAA;AAC3C,MAAA,MAAM,GAAA,GAAMd,4BAAUD,uBAAA,CAAM,CAAC,YAAY,QAAQ,CAAC,GAAG,IAAI,CAAA;AACzD,MAAA,MAAM,GAAA,GAAMC,4BAAUD,uBAAA,CAAM,CAAC,YAAY,UAAU,CAAC,GAAG,IAAI,CAAA;AAE3D,MAAA,IAAA,CAAK,SAAA,CAAU,KAAK,GAAG,CAAA;AACvB,MAAA,IAAA,CAAK,SAAA,CAAU,KAAK,GAAG,CAAA;AACvB,MAAA,IAAA,CAAK,kBAAkB,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA,EAAIE,2BAAA,CAAU,GAAG,CAAC,CAAA;AAC5D,MAAA,IAAA,CAAK,kBAAkB,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA,EAAIA,2BAAA,CAAU,GAAG,CAAC,CAAA;AAE5D,MAAA,IAAI,KAAK,IAAA,EAAM;AACb,QAAA,MAAM,GAAA,GAAMC,uBAAA,CAAM,CAAC,UAAU,CAAC,CAAA;AAC9B,QAAA,MAAM,GAAA,GAAMA,uBAAA,CAAM,CAAC,UAAU,CAAC,CAAA;AAC9B,QAAA,IAAA,CAAK,MAAA,CAAO,KAAK,GAAG,CAAA;AACpB,QAAA,IAAA,CAAK,MAAA,CAAO,KAAK,GAAG,CAAA;AACpB,QAAA,IAAA,CAAK,kBAAkB,CAAA,SAAA,EAAY,KAAK,CAAA,CAAA,EAAID,2BAAA,CAAU,GAAG,CAAC,CAAA;AAC1D,QAAA,IAAA,CAAK,kBAAkB,CAAA,SAAA,EAAY,KAAK,CAAA,CAAA,EAAIA,2BAAA,CAAU,GAAG,CAAC,CAAA;AAAA,MAC5D;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,WAAW,CAAA,EAAmB;AACpC,IAAA,OAAO,IAAA,CAAK,YAAA,KAAiB,MAAA,GAAS,IAAA,CAAK,IAAA,CAAK,CAAC,CAAA,GAAI,IAAA,CAAK,GAAA,CAAI,CAAA,EAAG,CAAC,CAAA;AAAA,EACpE;AAAA,EAEQ,GAAA,CAAI,OAAe,EAAA,EAA4C;AACrE,IAAA,iBAAA,CAAkB,OAAO,KAAK,CAAA;AAC9B,IAAA,MAAM,MAAA,GAAS,UAAA,CAAW,KAAA,EAAO,IAAA,CAAK,UAAU,CAAA;AAChD,IAAA,MAAM,EAAE,OAAO,MAAA,EAAQ,QAAA,EAAU,aAAa,WAAA,EAAa,SAAA,EAAW,YAAW,GAAI,MAAA;AAErF,IAAA,IAAI,QAAA,KAAa,KAAK,SAAA,EAAW;AAC/B,MAAA,MAAM,IAAIxB,4BAAA,CAAW,CAAA,oBAAA,EAAuB,KAAK,SAAS,CAAA,MAAA,EAAS,QAAQ,CAAA,CAAE,CAAA;AAAA,IAC/E;AACA,IAAA,IAAI,UAAU,CAAA,EAAG;AACf,MAAA,MAAM,IAAIK,uCAAA,CAAsB,kCAAA,EAAoC,QAAA,EAAU,MAAM,CAAA;AAAA,IACtF;AACA,IAAA,IAAI,CAAC,WAAA,IAAe,KAAA,IAAS,CAAA,EAAG;AAC9B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,6BAAA,EAA+B,OAAA,EAAS,KAAK,CAAA;AAAA,IAC/E;AAEA,IAAA,MAAM,CAAA,GAAI,WAAW,EAAA,EAAI,IAAA,CAAK,WAAW,KAAA,EAAO,IAAA,CAAK,UAAA,EAAY,WAAA,EAAa,IAAI,CAAA;AAClF,IAAA,MAAM,OAAA,GAAA,CAAW,WAAA,GAAc,MAAA,GAAS,KAAA,GAAQ,UAAU,IAAA,CAAK,UAAA;AAC/D,IAAA,MAAM,GAAA,GAAM,iBAAA,CAAkB,OAAA,EAAS,KAAA,CAAM,KAAK,CAAA;AAElD,IAAA,MAAM,QAAA,GAAW,IAAI,YAAA,CAAa,QAAQ,CAAA;AAE1C,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,MAAA,EAAQ,CAAA,EAAA,EAAK;AAC/B,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,KAAA,EAAO,CAAA,EAAA,EAAK;AAC9B,QAAA,MAAM,UAAA,GAAa,KAAA,CAAM,MAAA,GAAS,CAAA,GAAI,cAAc,CAAA,GAAI,SAAA;AACxD,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,QAAA,EAAU,CAAA,EAAA,EAAK;AACjC,UAAA,QAAA,CAAS,CAAC,CAAA,GAAI,WAAA,CAAY,KAAA,EAAO,UAAA,GAAa,IAAI,UAAU,CAAA;AAAA,QAC9D;AAEA,QAAA,IAAI,UAAA,GAAa,QAAA;AAEjB,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,WAAW,CAAA,EAAA,EAAK;AACvC,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,SAAA,CAAU,CAAC,CAAA;AAC5B,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,SAAA,CAAU,CAAC,CAAA;AAC5B,UAAA,IAAI,CAAC,GAAA,IAAO,CAAC,GAAA,EAAK;AAChB,YAAA,MAAM,IAAIL,6BAAW,qCAAqC,CAAA;AAAA,UAC5D;AAEA,UAAA,MAAM,YAAA,GAAe,CAAA,KAAM,CAAA,GAAI,IAAA,CAAK,YAAY,IAAA,CAAK,UAAA;AACrD,UAAA,MAAM,IAAA,GAAO,IAAI,YAAA,CAAa,IAAA,CAAK,UAAU,CAAA;AAC7C,UAAA,MAAM,MAAA,GAAS,EAAE,CAAC,CAAA;AAClB,UAAA,IAAI,CAAC,MAAA,EAAQ;AACX,YAAA,MAAM,IAAIA,6BAAW,0CAA0C,CAAA;AAAA,UACjE;AAEA,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC5B,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC5B,UAAA,MAAM,eAAe,MAAA,GAAU,MAAA,CAAO,OAAA,CAAQ,CAAC,KAAK,CAAA,GAAK,CAAA;AACzD,UAAA,MAAM,eAAe,MAAA,GAAU,MAAA,CAAO,OAAA,CAAQ,CAAC,KAAK,CAAA,GAAK,CAAA;AAEzD,UAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,YAAA,IAAIgD,IAAAA,GAAM,CAAA;AACV,YAAA,MAAM,OAAA,GAAU,GAAA,CAAI,MAAA,GAAS,CAAA,GAAI,UAAA;AACjC,YAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,YAAA,EAAc,CAAA,EAAA,EAAK;AACrC,cAAAA,IAAAA,IAAAA,CAAQ,WAAW,CAAC,CAAA,IAAK,KAAK,WAAA,CAAY,GAAA,EAAK,OAAA,GAAU,CAAA,GAAI,UAAU,CAAA;AAAA,YACzE;AAEA,YAAA,MAAM,OAAA,GAAU,GAAA,CAAI,MAAA,GAAS,CAAA,GAAI,UAAA;AACjC,YAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,cAAAA,IAAAA,IAAAA,CACG,MAAA,CAAO,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA,IAAK,WAAA,CAAY,GAAA,EAAK,OAAA,GAAU,CAAA,GAAI,UAAU,CAAA;AAAA,YACtF;AAEA,YAAA,IAAI,IAAA,CAAK,IAAA,IAAQ,MAAA,IAAU,MAAA,EAAQ;AACjC,cAAAA,QAAO,WAAA,CAAY,MAAA,EAAQ,MAAA,CAAO,MAAA,GAAS,IAAI,YAAY,CAAA;AAC3D,cAAAA,QAAO,WAAA,CAAY,MAAA,EAAQ,MAAA,CAAO,MAAA,GAAS,IAAI,YAAY,CAAA;AAAA,YAC7D;AAEA,YAAA,IAAA,CAAK,CAAC,CAAA,GAAI,IAAA,CAAK,UAAA,CAAWA,IAAG,CAAA;AAAA,UAC/B;AAEA,UAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,YAAA,MAAA,CAAO,IAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,GAAI,IAAA,CAAK,CAAC,CAAA,IAAK,CAAA;AAAA,UAC/C;AAEA,UAAA,UAAA,GAAa,IAAA;AAAA,QACf;AAEA,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,UAAA,MAAM,GAAA,GAAM,WAAA;AAAA,YACV,IAAA,CAAK,UAAA;AAAA,YACL,WAAA;AAAA,YACA,KAAA;AAAA,YACA,MAAA;AAAA,YACA,IAAA,CAAK,UAAA;AAAA,YACL,CAAA;AAAA,YACA,CAAA;AAAA,YACA;AAAA,WACF;AACA,UAAA,GAAA,CAAI,GAAG,CAAA,GAAI,UAAA,CAAW,CAAC,CAAA,IAAK,CAAA;AAAA,QAC9B;AAAA,MACF;AAAA,IACF;AAEA,IAAA,MAAM,WAAW,WAAA,GACb,CAAC,QAAQ,IAAA,CAAK,UAAU,IACxB,IAAA,CAAK,UAAA,GACH,CAAC,KAAA,EAAO,MAAA,EAAQ,KAAK,UAAU,CAAA,GAC/B,CAAC,MAAA,EAAQ,KAAA,EAAO,KAAK,UAAU,CAAA;AAErC,IAAA,OAAO;AAAA,MACL,MAAA,EAAQX,yBAAO,cAAA,CAAe;AAAA,QAC5B,IAAA,EAAM,GAAA;AAAA,QACN,KAAA,EAAO,QAAA;AAAA,QACP,OAAO,KAAA,CAAM,KAAA;AAAA,QACb,QAAQ,KAAA,CAAM;AAAA,OACf,CAAA;AAAA,MACD,CAAA,EAAG,SAAA;AAAA,QACD,CAAA;AAAA,QACA,IAAA,CAAK,SAAA;AAAA,QACL,KAAA;AAAA,QACA,IAAA,CAAK,UAAA;AAAA,QACL,KAAA,CAAM,KAAA;AAAA,QACN,KAAA,CAAM,MAAA;AAAA,QACN;AAAA;AACF,KACF;AAAA,EACF;AAAA,EAEA,WAAW,MAAA,EAA6B;AACtC,IAAA,IAAI,MAAA,CAAO,MAAA,GAAS,CAAA,IAAK,MAAA,CAAO,SAAS,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIhC,uCAAA,CAAsB,mCAAA,EAAqC,QAAA,EAAU,OAAO,MAAM,CAAA;AAAA,IAC9F;AACA,IAAA,MAAM,QAAA,GAAW,OAAO,CAAC,CAAA;AACzB,IAAA,IAAI,aAAa,MAAA,EAAW;AAC1B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,sCAAA,EAAwC,OAAA,EAAS,QAAQ,CAAA;AAAA,IAC3F;AACA,IAAA,MAAM,KAAA,GAAQ,aAAA,CAAc,QAAiB,CAAA;AAC7C,IAAA,MAAM,QAAQ,MAAA,CAAO,MAAA,KAAW,CAAA,GAAI,MAAA,CAAO,CAAC,CAAA,GAAI,MAAA;AAChD,IAAA,MAAM,KAAK,KAAA,KAAU,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,KAAW,CAAA;AACtE,IAAA,OAAO,IAAA,CAAK,GAAA,CAAI,KAAA,EAAO,EAAE,CAAA,CAAE,MAAA;AAAA,EAC7B;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,gBAAA,CAAiB,OAAkB,EAAA,EAAkC;AACnE,IAAA,MAAM,WAAA,GAAc,aAAA,CAAc,KAAc,CAAA;AAChD,IAAA,MAAM,WAAW,EAAA,KAAO,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,EAAQ,CAAA;AACtE,IAAA,MAAM,EAAE,MAAA,EAAQ,CAAA,KAAM,IAAA,CAAK,GAAA,CAAI,aAAa,QAAQ,CAAA;AACpD,IAAA,OAAO,CAAC,QAAQ,CAAC,CAAA;AAAA,EACnB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,IAAA,EAAO,KAAK,SAAS,CAAA,EAAA,EAAK,KAAK,UAAU,CAAA,aAAA,EAAgB,KAAK,SAAS,CAAA,CAAA,CAAA;AAAA,EAChF;AACF;AAiBO,IAAM,IAAA,GAAN,cAAmB,MAAA,CAAO;AAAA,EACd,SAAA;AAAA,EACA,UAAA;AAAA,EACA,SAAA;AAAA,EACA,IAAA;AAAA,EACA,UAAA;AAAA,EAET,SAAA;AAAA,EACA,SAAA;AAAA,EACA,MAAA;AAAA,EACA,MAAA;AAAA,EAER,WAAA,CACE,SAAA,EACA,UAAA,EACA,OAAA,GAII,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AACN,IAAA,mBAAA,CAAoB,aAAa,SAAS,CAAA;AAC1C,IAAA,mBAAA,CAAoB,cAAc,UAAU,CAAA;AAC5C,IAAA,MAAM,SAAA,GAAY,QAAQ,SAAA,IAAa,CAAA;AACvC,IAAA,mBAAA,CAAoB,aAAa,SAAS,CAAA;AAE1C,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,IAAA,GAAO,QAAQ,IAAA,IAAQ,IAAA;AAC5B,IAAA,IAAA,CAAK,UAAA,GAAa,QAAQ,UAAA,IAAc,IAAA;AAExC,IAAA,MAAM,IAAA,GAAO,CAAA,GAAM,IAAA,CAAK,IAAA,CAAK,UAAU,CAAA;AAEvC,IAAA,IAAA,CAAK,YAAY,EAAC;AAClB,IAAA,IAAA,CAAK,YAAY,EAAC;AAClB,IAAA,IAAA,CAAK,SAAS,EAAC;AACf,IAAA,IAAA,CAAK,SAAS,EAAC;AAEf,IAAA,KAAA,IAAS,KAAA,GAAQ,CAAA,EAAG,KAAA,GAAQ,IAAA,CAAK,WAAW,KAAA,EAAA,EAAS;AACnD,MAAA,MAAM,QAAA,GAAW,KAAA,KAAU,CAAA,GAAI,SAAA,GAAY,UAAA;AAE3C,MAAA,MAAM,GAAA,GAAMkB,4BAAUD,uBAAA,CAAM,CAAC,IAAI,UAAA,EAAY,QAAQ,CAAC,CAAA,EAAG,IAAI,CAAA;AAC7D,MAAA,MAAM,GAAA,GAAMC,4BAAUD,uBAAA,CAAM,CAAC,IAAI,UAAA,EAAY,UAAU,CAAC,CAAA,EAAG,IAAI,CAAA;AAE/D,MAAA,IAAA,CAAK,SAAA,CAAU,KAAK,GAAG,CAAA;AACvB,MAAA,IAAA,CAAK,SAAA,CAAU,KAAK,GAAG,CAAA;AACvB,MAAA,IAAA,CAAK,kBAAkB,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA,EAAIE,2BAAA,CAAU,GAAG,CAAC,CAAA;AAC5D,MAAA,IAAA,CAAK,kBAAkB,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA,EAAIA,2BAAA,CAAU,GAAG,CAAC,CAAA;AAE5D,MAAA,IAAI,KAAK,IAAA,EAAM;AACb,QAAA,MAAM,GAAA,GAAMC,uBAAA,CAAM,CAAC,CAAA,GAAI,UAAU,CAAC,CAAA;AAClC,QAAA,MAAM,GAAA,GAAMA,uBAAA,CAAM,CAAC,CAAA,GAAI,UAAU,CAAC,CAAA;AAClC,QAAA,IAAA,CAAK,MAAA,CAAO,KAAK,GAAG,CAAA;AACpB,QAAA,IAAA,CAAK,MAAA,CAAO,KAAK,GAAG,CAAA;AACpB,QAAA,IAAA,CAAK,kBAAkB,CAAA,SAAA,EAAY,KAAK,CAAA,CAAA,EAAID,2BAAA,CAAU,GAAG,CAAC,CAAA;AAC1D,QAAA,IAAA,CAAK,kBAAkB,CAAA,SAAA,EAAY,KAAK,CAAA,CAAA,EAAIA,2BAAA,CAAU,GAAG,CAAC,CAAA;AAAA,MAC5D;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,QAAQ,CAAA,EAAmB;AACjC,IAAA,OAAO,CAAA,IAAK,CAAA,GAAI,IAAA,CAAK,GAAA,CAAI,CAAC,CAAC,CAAA,CAAA;AAAA,EAC7B;AAAA,EAEQ,GAAA,CAAI,KAAA,EAAe,EAAA,EAAa,EAAA,EAAuD;AAC7F,IAAA,iBAAA,CAAkB,OAAO,MAAM,CAAA;AAC/B,IAAA,MAAM,MAAA,GAAS,UAAA,CAAW,KAAA,EAAO,IAAA,CAAK,UAAU,CAAA;AAChD,IAAA,MAAM,EAAE,OAAO,MAAA,EAAQ,QAAA,EAAU,aAAa,WAAA,EAAa,SAAA,EAAW,YAAW,GAAI,MAAA;AAErF,IAAA,IAAI,QAAA,KAAa,KAAK,SAAA,EAAW;AAC/B,MAAA,MAAM,IAAIxB,4BAAA,CAAW,CAAA,oBAAA,EAAuB,KAAK,SAAS,CAAA,MAAA,EAAS,QAAQ,CAAA,CAAE,CAAA;AAAA,IAC/E;AACA,IAAA,IAAI,UAAU,CAAA,EAAG;AACf,MAAA,MAAM,IAAIK,uCAAA,CAAsB,kCAAA,EAAoC,QAAA,EAAU,MAAM,CAAA;AAAA,IACtF;AACA,IAAA,IAAI,CAAC,WAAA,IAAe,KAAA,IAAS,CAAA,EAAG;AAC9B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,6BAAA,EAA+B,OAAA,EAAS,KAAK,CAAA;AAAA,IAC/E;AAEA,IAAA,MAAM,CAAA,GAAI,WAAW,EAAA,EAAI,IAAA,CAAK,WAAW,KAAA,EAAO,IAAA,CAAK,UAAA,EAAY,WAAA,EAAa,IAAI,CAAA;AAClF,IAAA,MAAM,CAAA,GAAI,WAAW,EAAA,EAAI,IAAA,CAAK,WAAW,KAAA,EAAO,IAAA,CAAK,UAAA,EAAY,WAAA,EAAa,IAAI,CAAA;AAElF,IAAA,MAAM,OAAA,GAAA,CAAW,WAAA,GAAc,MAAA,GAAS,KAAA,GAAQ,UAAU,IAAA,CAAK,UAAA;AAC/D,IAAA,MAAM,GAAA,GAAM,iBAAA,CAAkB,OAAA,EAAS,KAAA,CAAM,KAAK,CAAA;AAElD,IAAA,MAAM,QAAA,GAAW,IAAI,YAAA,CAAa,QAAQ,CAAA;AAC1C,IAAA,MAAM,KAAA,GAAQ,IAAI,YAAA,CAAa,CAAA,GAAI,KAAK,UAAU,CAAA;AAElD,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,MAAA,EAAQ,CAAA,EAAA,EAAK;AAC/B,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,KAAA,EAAO,CAAA,EAAA,EAAK;AAC9B,QAAA,MAAM,UAAA,GAAa,KAAA,CAAM,MAAA,GAAS,CAAA,GAAI,cAAc,CAAA,GAAI,SAAA;AACxD,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,QAAA,EAAU,CAAA,EAAA,EAAK;AACjC,UAAA,QAAA,CAAS,CAAC,CAAA,GAAI,WAAA,CAAY,KAAA,EAAO,UAAA,GAAa,IAAI,UAAU,CAAA;AAAA,QAC9D;AAEA,QAAA,IAAI,UAAA,GAAa,QAAA;AAEjB,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,WAAW,CAAA,EAAA,EAAK;AACvC,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,SAAA,CAAU,CAAC,CAAA;AAC5B,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,SAAA,CAAU,CAAC,CAAA;AAC5B,UAAA,IAAI,CAAC,GAAA,IAAO,CAAC,GAAA,EAAK;AAChB,YAAA,MAAM,IAAIL,6BAAW,sCAAsC,CAAA;AAAA,UAC7D;AAEA,UAAA,MAAM,YAAA,GAAe,CAAA,KAAM,CAAA,GAAI,IAAA,CAAK,YAAY,IAAA,CAAK,UAAA;AACrD,UAAA,MAAM,MAAA,GAAS,EAAE,CAAC,CAAA;AAClB,UAAA,MAAM,MAAA,GAAS,EAAE,CAAC,CAAA;AAClB,UAAA,IAAI,CAAC,MAAA,IAAU,CAAC,MAAA,EAAQ;AACtB,YAAA,MAAM,IAAIA,6BAAW,oCAAoC,CAAA;AAAA,UAC3D;AAEA,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC5B,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC5B,UAAA,MAAM,eAAe,MAAA,GAAU,MAAA,CAAO,OAAA,CAAQ,CAAC,KAAK,CAAA,GAAK,CAAA;AACzD,UAAA,MAAM,eAAe,MAAA,GAAU,MAAA,CAAO,OAAA,CAAQ,CAAC,KAAK,CAAA,GAAK,CAAA;AAEzD,UAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AAC5C,YAAA,IAAIgD,IAAAA,GAAM,CAAA;AACV,YAAA,MAAM,OAAA,GAAU,GAAA,CAAI,MAAA,GAAS,CAAA,GAAI,UAAA;AACjC,YAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,YAAA,EAAc,CAAA,EAAA,EAAK;AACrC,cAAAA,IAAAA,IAAAA,CAAQ,WAAW,CAAC,CAAA,IAAK,KAAK,WAAA,CAAY,GAAA,EAAK,OAAA,GAAU,CAAA,GAAI,UAAU,CAAA;AAAA,YACzE;AACA,YAAA,MAAM,OAAA,GAAU,GAAA,CAAI,MAAA,GAAS,CAAA,GAAI,UAAA;AACjC,YAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,cAAAA,IAAAA,IAAAA,CACG,MAAA,CAAO,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA,IAAK,WAAA,CAAY,GAAA,EAAK,OAAA,GAAU,CAAA,GAAI,UAAU,CAAA;AAAA,YACtF;AACA,YAAA,IAAI,IAAA,CAAK,IAAA,IAAQ,MAAA,IAAU,MAAA,EAAQ;AACjC,cAAAA,QAAO,WAAA,CAAY,MAAA,EAAQ,MAAA,CAAO,MAAA,GAAS,IAAI,YAAY,CAAA;AAC3D,cAAAA,QAAO,WAAA,CAAY,MAAA,EAAQ,MAAA,CAAO,MAAA,GAAS,IAAI,YAAY,CAAA;AAAA,YAC7D;AACA,YAAA,KAAA,CAAM,CAAC,CAAA,GAAIA,IAAAA;AAAA,UACb;AAEA,UAAA,MAAM,IAAA,GAAO,IAAI,YAAA,CAAa,IAAA,CAAK,UAAU,CAAA;AAC7C,UAAA,MAAM,IAAA,GAAO,IAAI,YAAA,CAAa,IAAA,CAAK,UAAU,CAAA;AAE7C,UAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,YAAA,MAAM,QAAQ,IAAA,CAAK,OAAA,CAAQ,KAAA,CAAM,CAAC,KAAK,CAAC,CAAA;AACxC,YAAA,MAAM,KAAA,GAAQ,KAAK,OAAA,CAAQ,KAAA,CAAM,KAAK,UAAA,GAAa,CAAC,KAAK,CAAC,CAAA;AAC1D,YAAA,MAAM,KAAA,GAAQ,KAAK,IAAA,CAAK,KAAA,CAAM,IAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAC,CAAA;AAC3D,YAAA,MAAM,KAAA,GAAQ,KAAK,OAAA,CAAQ,KAAA,CAAM,IAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAC,CAAA;AAE9D,YAAA,MAAM,QAAQ,MAAA,CAAO,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA;AACjD,YAAA,MAAM,KAAA,GAAQ,KAAA,GAAQ,KAAA,GAAQ,KAAA,GAAQ,KAAA;AACtC,YAAA,MAAM,KAAA,GAAQ,KAAA,GAAQ,IAAA,CAAK,IAAA,CAAK,KAAK,CAAA;AAErC,YAAA,IAAA,CAAK,CAAC,CAAA,GAAI,KAAA;AACV,YAAA,IAAA,CAAK,CAAC,CAAA,GAAI,KAAA;AAAA,UACZ;AAEA,UAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,YAAA,MAAA,CAAO,IAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,GAAI,IAAA,CAAK,CAAC,CAAA,IAAK,CAAA;AAC7C,YAAA,MAAA,CAAO,IAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,GAAI,IAAA,CAAK,CAAC,CAAA,IAAK,CAAA;AAAA,UAC/C;AAEA,UAAA,UAAA,GAAa,IAAA;AAAA,QACf;AAEA,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,UAAA,MAAM,GAAA,GAAM,WAAA;AAAA,YACV,IAAA,CAAK,UAAA;AAAA,YACL,WAAA;AAAA,YACA,KAAA;AAAA,YACA,MAAA;AAAA,YACA,IAAA,CAAK,UAAA;AAAA,YACL,CAAA;AAAA,YACA,CAAA;AAAA,YACA;AAAA,WACF;AACA,UAAA,GAAA,CAAI,GAAG,CAAA,GAAI,UAAA,CAAW,CAAC,CAAA,IAAK,CAAA;AAAA,QAC9B;AAAA,MACF;AAAA,IACF;AAEA,IAAA,MAAM,WAAW,WAAA,GACb,CAAC,QAAQ,IAAA,CAAK,UAAU,IACxB,IAAA,CAAK,UAAA,GACH,CAAC,KAAA,EAAO,MAAA,EAAQ,KAAK,UAAU,CAAA,GAC/B,CAAC,MAAA,EAAQ,KAAA,EAAO,KAAK,UAAU,CAAA;AAErC,IAAA,OAAO;AAAA,MACL,MAAA,EAAQX,yBAAO,cAAA,CAAe;AAAA,QAC5B,IAAA,EAAM,GAAA;AAAA,QACN,KAAA,EAAO,QAAA;AAAA,QACP,OAAO,KAAA,CAAM,KAAA;AAAA,QACb,QAAQ,KAAA,CAAM;AAAA,OACf,CAAA;AAAA,MACD,CAAA,EAAG,SAAA;AAAA,QACD,CAAA;AAAA,QACA,IAAA,CAAK,SAAA;AAAA,QACL,KAAA;AAAA,QACA,IAAA,CAAK,UAAA;AAAA,QACL,KAAA,CAAM,KAAA;AAAA,QACN,KAAA,CAAM,MAAA;AAAA,QACN;AAAA,OACF;AAAA,MACA,CAAA,EAAG,SAAA;AAAA,QACD,CAAA;AAAA,QACA,IAAA,CAAK,SAAA;AAAA,QACL,KAAA;AAAA,QACA,IAAA,CAAK,UAAA;AAAA,QACL,KAAA,CAAM,KAAA;AAAA,QACN,KAAA,CAAM,MAAA;AAAA,QACN;AAAA;AACF,KACF;AAAA,EACF;AAAA,EAEA,WAAW,MAAA,EAA6B;AACtC,IAAA,IAAI,MAAA,CAAO,MAAA,GAAS,CAAA,IAAK,MAAA,CAAO,SAAS,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIhC,uCAAA;AAAA,QACR,oCAAA;AAAA,QACA,QAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAAA,IACF;AACA,IAAA,MAAM,QAAA,GAAW,OAAO,CAAC,CAAA;AACzB,IAAA,IAAI,aAAa,MAAA,EAAW;AAC1B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,uCAAA,EAAyC,OAAA,EAAS,QAAQ,CAAA;AAAA,IAC5F;AACA,IAAA,MAAM,KAAA,GAAQ,aAAA,CAAc,QAAiB,CAAA;AAC7C,IAAA,MAAM,QAAQ,MAAA,CAAO,MAAA,IAAU,CAAA,GAAI,MAAA,CAAO,CAAC,CAAA,GAAI,MAAA;AAC/C,IAAA,MAAM,QAAQ,MAAA,CAAO,MAAA,IAAU,CAAA,GAAI,MAAA,CAAO,CAAC,CAAA,GAAI,MAAA;AAC/C,IAAA,MAAM,KAAK,KAAA,KAAU,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,KAAW,CAAA;AACtE,IAAA,MAAM,KAAK,KAAA,KAAU,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,KAAW,CAAA;AACtE,IAAA,OAAO,IAAA,CAAK,GAAA,CAAI,KAAA,EAAO,EAAA,EAAI,EAAE,CAAA,CAAE,MAAA;AAAA,EACjC;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,gBAAA,CAAiB,KAAA,EAAkB,EAAA,EAAgB,EAAA,EAA4C;AAC7F,IAAA,MAAM,WAAA,GAAc,aAAA,CAAc,KAAc,CAAA;AAChD,IAAA,MAAM,WAAW,EAAA,KAAO,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,EAAQ,CAAA;AACtE,IAAA,MAAM,WAAW,EAAA,KAAO,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,EAAQ,CAAA;AACtE,IAAA,MAAM,EAAE,QAAQ,CAAA,EAAG,CAAA,KAAM,IAAA,CAAK,GAAA,CAAI,WAAA,EAAa,QAAA,EAAU,QAAQ,CAAA;AACjE,IAAA,OAAO,CAAC,MAAA,EAAQ,CAAC,CAAA,EAAG,CAAC,CAAC,CAAA;AAAA,EACxB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,KAAA,EAAQ,KAAK,SAAS,CAAA,EAAA,EAAK,KAAK,UAAU,CAAA,aAAA,EAAgB,KAAK,SAAS,CAAA,CAAA,CAAA;AAAA,EACjF;AACF;AAeO,IAAM,GAAA,GAAN,cAAkB,MAAA,CAAO;AAAA,EACb,SAAA;AAAA,EACA,UAAA;AAAA,EACA,SAAA;AAAA,EACA,IAAA;AAAA,EACA,UAAA;AAAA,EAET,SAAA;AAAA,EACA,SAAA;AAAA,EACA,MAAA;AAAA,EACA,MAAA;AAAA,EAER,WAAA,CACE,SAAA,EACA,UAAA,EACA,OAAA,GAII,EAAC,EACL;AACA,IAAA,KAAA,EAAM;AACN,IAAA,mBAAA,CAAoB,aAAa,SAAS,CAAA;AAC1C,IAAA,mBAAA,CAAoB,cAAc,UAAU,CAAA;AAC5C,IAAA,MAAM,SAAA,GAAY,QAAQ,SAAA,IAAa,CAAA;AACvC,IAAA,mBAAA,CAAoB,aAAa,SAAS,CAAA;AAE1C,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,IAAA,GAAO,QAAQ,IAAA,IAAQ,IAAA;AAC5B,IAAA,IAAA,CAAK,UAAA,GAAa,QAAQ,UAAA,IAAc,IAAA;AAExC,IAAA,MAAM,IAAA,GAAO,CAAA,GAAM,IAAA,CAAK,IAAA,CAAK,UAAU,CAAA;AAEvC,IAAA,IAAA,CAAK,YAAY,EAAC;AAClB,IAAA,IAAA,CAAK,YAAY,EAAC;AAClB,IAAA,IAAA,CAAK,SAAS,EAAC;AACf,IAAA,IAAA,CAAK,SAAS,EAAC;AAEf,IAAA,KAAA,IAAS,KAAA,GAAQ,CAAA,EAAG,KAAA,GAAQ,IAAA,CAAK,WAAW,KAAA,EAAA,EAAS;AACnD,MAAA,MAAM,QAAA,GAAW,KAAA,KAAU,CAAA,GAAI,SAAA,GAAY,UAAA;AAE3C,MAAA,MAAM,GAAA,GAAMkB,4BAAUD,uBAAA,CAAM,CAAC,IAAI,UAAA,EAAY,QAAQ,CAAC,CAAA,EAAG,IAAI,CAAA;AAC7D,MAAA,MAAM,GAAA,GAAMC,4BAAUD,uBAAA,CAAM,CAAC,IAAI,UAAA,EAAY,UAAU,CAAC,CAAA,EAAG,IAAI,CAAA;AAE/D,MAAA,IAAA,CAAK,SAAA,CAAU,KAAK,GAAG,CAAA;AACvB,MAAA,IAAA,CAAK,SAAA,CAAU,KAAK,GAAG,CAAA;AACvB,MAAA,IAAA,CAAK,kBAAkB,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA,EAAIE,2BAAA,CAAU,GAAG,CAAC,CAAA;AAC5D,MAAA,IAAA,CAAK,kBAAkB,CAAA,WAAA,EAAc,KAAK,CAAA,CAAA,EAAIA,2BAAA,CAAU,GAAG,CAAC,CAAA;AAE5D,MAAA,IAAI,KAAK,IAAA,EAAM;AACb,QAAA,MAAM,GAAA,GAAMC,uBAAA,CAAM,CAAC,CAAA,GAAI,UAAU,CAAC,CAAA;AAClC,QAAA,MAAM,GAAA,GAAMA,uBAAA,CAAM,CAAC,CAAA,GAAI,UAAU,CAAC,CAAA;AAClC,QAAA,IAAA,CAAK,MAAA,CAAO,KAAK,GAAG,CAAA;AACpB,QAAA,IAAA,CAAK,MAAA,CAAO,KAAK,GAAG,CAAA;AACpB,QAAA,IAAA,CAAK,kBAAkB,CAAA,SAAA,EAAY,KAAK,CAAA,CAAA,EAAID,2BAAA,CAAU,GAAG,CAAC,CAAA;AAC1D,QAAA,IAAA,CAAK,kBAAkB,CAAA,SAAA,EAAY,KAAK,CAAA,CAAA,EAAIA,2BAAA,CAAU,GAAG,CAAC,CAAA;AAAA,MAC5D;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,QAAQ,CAAA,EAAmB;AACjC,IAAA,OAAO,CAAA,IAAK,CAAA,GAAI,IAAA,CAAK,GAAA,CAAI,CAAC,CAAC,CAAA,CAAA;AAAA,EAC7B;AAAA,EAEQ,GAAA,CAAI,OAAe,EAAA,EAA4C;AACrE,IAAA,iBAAA,CAAkB,OAAO,KAAK,CAAA;AAC9B,IAAA,MAAM,MAAA,GAAS,UAAA,CAAW,KAAA,EAAO,IAAA,CAAK,UAAU,CAAA;AAChD,IAAA,MAAM,EAAE,OAAO,MAAA,EAAQ,QAAA,EAAU,aAAa,WAAA,EAAa,SAAA,EAAW,YAAW,GAAI,MAAA;AAErF,IAAA,IAAI,QAAA,KAAa,KAAK,SAAA,EAAW;AAC/B,MAAA,MAAM,IAAIxB,4BAAA,CAAW,CAAA,oBAAA,EAAuB,KAAK,SAAS,CAAA,MAAA,EAAS,QAAQ,CAAA,CAAE,CAAA;AAAA,IAC/E;AACA,IAAA,IAAI,UAAU,CAAA,EAAG;AACf,MAAA,MAAM,IAAIK,uCAAA,CAAsB,kCAAA,EAAoC,QAAA,EAAU,MAAM,CAAA;AAAA,IACtF;AACA,IAAA,IAAI,CAAC,WAAA,IAAe,KAAA,IAAS,CAAA,EAAG;AAC9B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,6BAAA,EAA+B,OAAA,EAAS,KAAK,CAAA;AAAA,IAC/E;AAEA,IAAA,MAAM,CAAA,GAAI,WAAW,EAAA,EAAI,IAAA,CAAK,WAAW,KAAA,EAAO,IAAA,CAAK,UAAA,EAAY,WAAA,EAAa,IAAI,CAAA;AAClF,IAAA,MAAM,OAAA,GAAA,CAAW,WAAA,GAAc,MAAA,GAAS,KAAA,GAAQ,UAAU,IAAA,CAAK,UAAA;AAC/D,IAAA,MAAM,GAAA,GAAM,iBAAA,CAAkB,OAAA,EAAS,KAAA,CAAM,KAAK,CAAA;AAElD,IAAA,MAAM,QAAA,GAAW,IAAI,YAAA,CAAa,QAAQ,CAAA;AAC1C,IAAA,MAAM,OAAA,GAAU,IAAI,YAAA,CAAa,CAAA,GAAI,KAAK,UAAU,CAAA;AACpD,IAAA,MAAM,OAAA,GAAU,IAAI,YAAA,CAAa,CAAA,GAAI,KAAK,UAAU,CAAA;AAEpD,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,MAAA,EAAQ,CAAA,EAAA,EAAK;AAC/B,MAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,KAAA,EAAO,CAAA,EAAA,EAAK;AAC9B,QAAA,MAAM,UAAA,GAAa,KAAA,CAAM,MAAA,GAAS,CAAA,GAAI,cAAc,CAAA,GAAI,SAAA;AACxD,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,QAAA,EAAU,CAAA,EAAA,EAAK;AACjC,UAAA,QAAA,CAAS,CAAC,CAAA,GAAI,WAAA,CAAY,KAAA,EAAO,UAAA,GAAa,IAAI,UAAU,CAAA;AAAA,QAC9D;AAEA,QAAA,IAAI,UAAA,GAAa,QAAA;AAEjB,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,WAAW,CAAA,EAAA,EAAK;AACvC,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,SAAA,CAAU,CAAC,CAAA;AAC5B,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,SAAA,CAAU,CAAC,CAAA;AAC5B,UAAA,IAAI,CAAC,GAAA,IAAO,CAAC,GAAA,EAAK;AAChB,YAAA,MAAM,IAAIL,6BAAW,qCAAqC,CAAA;AAAA,UAC5D;AAEA,UAAA,MAAM,YAAA,GAAe,CAAA,KAAM,CAAA,GAAI,IAAA,CAAK,YAAY,IAAA,CAAK,UAAA;AACrD,UAAA,MAAM,MAAA,GAAS,EAAE,CAAC,CAAA;AAClB,UAAA,IAAI,CAAC,MAAA,EAAQ;AACX,YAAA,MAAM,IAAIA,6BAAW,0CAA0C,CAAA;AAAA,UACjE;AAEA,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,UAAA,GAAa,GAAA,CAAI,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACrC,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC5B,UAAA,MAAM,MAAA,GAAS,IAAA,CAAK,MAAA,CAAO,CAAC,CAAA;AAC5B,UAAA,MAAM,eAAe,MAAA,GAAU,MAAA,CAAO,OAAA,CAAQ,CAAC,KAAK,CAAA,GAAK,CAAA;AACzD,UAAA,MAAM,eAAe,MAAA,GAAU,MAAA,CAAO,OAAA,CAAQ,CAAC,KAAK,CAAA,GAAK,CAAA;AAEzD,UAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AAC5C,YAAA,IAAI,KAAA,GAAQ,CAAA;AACZ,YAAA,IAAI,KAAA,GAAQ,CAAA;AACZ,YAAA,MAAM,OAAA,GAAU,GAAA,CAAI,MAAA,GAAS,CAAA,GAAI,UAAA;AACjC,YAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,YAAA,EAAc,CAAA,EAAA,EAAK;AACrC,cAAA,KAAA,IAAA,CAAU,UAAA,CAAW,CAAC,CAAA,IAAK,CAAA,IAAK,YAAY,GAAA,EAAK,OAAA,GAAU,IAAI,UAAU,CAAA;AAAA,YAC3E;AACA,YAAA,MAAM,OAAA,GAAU,GAAA,CAAI,MAAA,GAAS,CAAA,GAAI,UAAA;AACjC,YAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,cAAA,KAAA,IAAA,CACG,MAAA,CAAO,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA,IAAK,WAAA,CAAY,GAAA,EAAK,OAAA,GAAU,CAAA,GAAI,UAAU,CAAA;AAAA,YACtF;AACA,YAAA,IAAI,IAAA,CAAK,IAAA,IAAQ,MAAA,IAAU,MAAA,EAAQ;AACjC,cAAA,KAAA,IAAS,WAAA,CAAY,MAAA,EAAQ,MAAA,CAAO,MAAA,GAAS,IAAI,YAAY,CAAA;AAC7D,cAAA,KAAA,IAAS,WAAA,CAAY,MAAA,EAAQ,MAAA,CAAO,MAAA,GAAS,IAAI,YAAY,CAAA;AAAA,YAC/D;AACA,YAAA,OAAA,CAAQ,CAAC,CAAA,GAAI,KAAA;AACb,YAAA,OAAA,CAAQ,CAAC,CAAA,GAAI,KAAA;AAAA,UACf;AAEA,UAAA,MAAM,IAAA,GAAO,IAAI,YAAA,CAAa,IAAA,CAAK,UAAU,CAAA;AAC7C,UAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,YAAA,MAAM,CAAA,GAAI,IAAA,CAAK,OAAA,CAAA,CAAS,OAAA,CAAQ,CAAC,KAAK,CAAA,KAAM,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA,CAAE,CAAA;AAC5D,YAAA,MAAM,IAAI,IAAA,CAAK,OAAA;AAAA,cAAA,CACZ,OAAA,CAAQ,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,MAAM,OAAA,CAAQ,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA;AAAA,aACzE;AACA,YAAA,MAAM,IAAI,IAAA,CAAK,IAAA;AAAA,cAAA,CACZ,OAAA,CAAQ,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA,IAAK,CAAA,IAAK,OAAA,CAAQ,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA;AAAA,aACrF;AACA,YAAA,IAAA,CAAK,CAAC,CAAA,GAAA,CAAK,CAAA,GAAI,CAAA,IAAK,CAAA,GAAI,CAAA,IAAK,MAAA,CAAO,CAAA,GAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,IAAK,CAAA,CAAA;AAAA,UAClE;AAEA,UAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,YAAA,MAAA,CAAO,IAAI,IAAA,CAAK,UAAA,GAAa,CAAC,CAAA,GAAI,IAAA,CAAK,CAAC,CAAA,IAAK,CAAA;AAAA,UAC/C;AAEA,UAAA,UAAA,GAAa,IAAA;AAAA,QACf;AAEA,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,YAAY,CAAA,EAAA,EAAK;AACxC,UAAA,MAAM,GAAA,GAAM,WAAA;AAAA,YACV,IAAA,CAAK,UAAA;AAAA,YACL,WAAA;AAAA,YACA,KAAA;AAAA,YACA,MAAA;AAAA,YACA,IAAA,CAAK,UAAA;AAAA,YACL,CAAA;AAAA,YACA,CAAA;AAAA,YACA;AAAA,WACF;AACA,UAAA,GAAA,CAAI,GAAG,CAAA,GAAI,UAAA,CAAW,CAAC,CAAA,IAAK,CAAA;AAAA,QAC9B;AAAA,MACF;AAAA,IACF;AAEA,IAAA,MAAM,WAAW,WAAA,GACb,CAAC,QAAQ,IAAA,CAAK,UAAU,IACxB,IAAA,CAAK,UAAA,GACH,CAAC,KAAA,EAAO,MAAA,EAAQ,KAAK,UAAU,CAAA,GAC/B,CAAC,MAAA,EAAQ,KAAA,EAAO,KAAK,UAAU,CAAA;AAErC,IAAA,OAAO;AAAA,MACL,MAAA,EAAQqC,yBAAO,cAAA,CAAe;AAAA,QAC5B,IAAA,EAAM,GAAA;AAAA,QACN,KAAA,EAAO,QAAA;AAAA,QACP,OAAO,KAAA,CAAM,KAAA;AAAA,QACb,QAAQ,KAAA,CAAM;AAAA,OACf,CAAA;AAAA,MACD,CAAA,EAAG,SAAA;AAAA,QACD,CAAA;AAAA,QACA,IAAA,CAAK,SAAA;AAAA,QACL,KAAA;AAAA,QACA,IAAA,CAAK,UAAA;AAAA,QACL,KAAA,CAAM,KAAA;AAAA,QACN,KAAA,CAAM,MAAA;AAAA,QACN;AAAA;AACF,KACF;AAAA,EACF;AAAA,EAEA,WAAW,MAAA,EAA6B;AACtC,IAAA,IAAI,MAAA,CAAO,MAAA,GAAS,CAAA,IAAK,MAAA,CAAO,SAAS,CAAA,EAAG;AAC1C,MAAA,MAAM,IAAIhC,uCAAA,CAAsB,mCAAA,EAAqC,QAAA,EAAU,OAAO,MAAM,CAAA;AAAA,IAC9F;AACA,IAAA,MAAM,QAAA,GAAW,OAAO,CAAC,CAAA;AACzB,IAAA,IAAI,aAAa,MAAA,EAAW;AAC1B,MAAA,MAAM,IAAIA,uCAAA,CAAsB,sCAAA,EAAwC,OAAA,EAAS,QAAQ,CAAA;AAAA,IAC3F;AACA,IAAA,MAAM,KAAA,GAAQ,aAAA,CAAc,QAAiB,CAAA;AAC7C,IAAA,MAAM,QAAQ,MAAA,CAAO,MAAA,KAAW,CAAA,GAAI,MAAA,CAAO,CAAC,CAAA,GAAI,MAAA;AAChD,IAAA,MAAM,KAAK,KAAA,KAAU,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,KAAW,CAAA;AACtE,IAAA,OAAO,IAAA,CAAK,GAAA,CAAI,KAAA,EAAO,EAAE,CAAA,CAAE,MAAA;AAAA,EAC7B;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,gBAAA,CAAiB,OAAkB,EAAA,EAAkC;AACnE,IAAA,MAAM,WAAA,GAAc,aAAA,CAAc,KAAc,CAAA;AAChD,IAAA,MAAM,WAAW,EAAA,KAAO,MAAA,GAAY,MAAA,GAAY,aAAA,CAAc,EAAQ,CAAA;AACtE,IAAA,MAAM,EAAE,MAAA,EAAQ,CAAA,KAAM,IAAA,CAAK,GAAA,CAAI,aAAa,QAAQ,CAAA;AACpD,IAAA,OAAO,CAAC,QAAQ,CAAC,CAAA;AAAA,EACnB;AAAA,EAES,QAAA,GAAmB;AAC1B,IAAA,OAAO,CAAA,IAAA,EAAO,KAAK,SAAS,CAAA,EAAA,EAAK,KAAK,UAAU,CAAA,aAAA,EAAgB,KAAK,SAAS,CAAA,CAAA,CAAA;AAAA,EAChF;AACF;;;ACt+BA,SAAS,QAAA,CAAS,SAAiB,UAAA,EAA4B;AAC7D,EAAA,MAAM,WAAW,OAAA,CAAQ,IAAA;AACzB,EAAA,MAAM,OAAA,GAAU,IAAI,YAAA,CAAa,QAAA,GAAW,UAAU,CAAA;AAEtD,EAAA,MAAM,OAAO,OAAA,CAAQ,IAAA;AACrB,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,MAAM,IAAIH,6BAAW,iDAAiD,CAAA;AAAA,EACxE;AAEA,EAAA,MAAM,OAAA,GAAU,OAAA,CAAQ,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACtC,EAAA,MAAM,OAAO,OAAA,CAAQ,MAAA;AAErB,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,QAAA,EAAU,CAAA,EAAA,EAAK;AACjC,IAAA,MAAM,MAAA,GAAS,OAAO,CAAA,GAAI,OAAA;AAC1B,IAAA,IAAI,GAAA;AACJ,IAAA,IAAI,gBAAgB,aAAA,EAAe;AACjC,MAAA,MAAM,GAAA,GAAMiC,kCAAA,CAAiB,IAAA,EAAM,MAAM,CAAA;AACzC,MAAA,MAAM,QAAA,GAAW,OAAO,GAAG,CAAA;AAC3B,MAAA,IAAI,CAAC,MAAA,CAAO,aAAA,CAAc,QAAQ,CAAA,EAAG;AACnC,QAAA,MAAM,IAAI9B,uCAAA;AAAA,UACR,CAAA,YAAA,EAAe,GAAA,CAAI,QAAA,EAAU,CAAA,2BAAA,CAAA;AAAA,UAC7B,QAAA;AAAA,UACA,IAAI,QAAA;AAAS,SACf;AAAA,MACF;AACA,MAAA,GAAA,GAAM,QAAA;AAAA,IACR,CAAA,MAAO;AACL,MAAA,GAAA,GAAM,MAAA,CAAO+B,mCAAA,CAAkB,IAAA,EAAM,MAAM,CAAC,CAAA;AAAA,IAC9C;AAEA,IAAA,IAAI,CAAC,OAAO,QAAA,CAAS,GAAG,KAAK,CAAC,MAAA,CAAO,SAAA,CAAU,GAAG,CAAA,EAAG;AACnD,MAAA,MAAM,IAAI/B,uCAAA,CAAsB,CAAA,YAAA,EAAe,GAAG,CAAA,uBAAA,CAAA,EAA2B,UAAU,GAAG,CAAA;AAAA,IAC5F;AAEA,IAAA,IAAI,GAAA,GAAM,CAAA,IAAK,GAAA,IAAO,UAAA,EAAY;AAChC,MAAA,MAAM,IAAIA,uCAAA;AAAA,QACR,CAAA,YAAA,EAAe,GAAG,CAAA,kBAAA,EAAqB,UAAU,CAAA,CAAA,CAAA;AAAA,QACjD,QAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAA,CAAQ,CAAA,GAAI,UAAA,GAAa,GAAG,CAAA,GAAI,CAAA;AAAA,EAClC;AAEA,EAAA,OAAOgC,yBAAO,cAAA,CAAe;AAAA,IAC3B,IAAA,EAAM,OAAA;AAAA,IACN,KAAA,EAAO,CAAC,QAAA,EAAU,UAAU,CAAA;AAAA,IAC5B,KAAA,EAAO,SAAA;AAAA,IACP,QAAQ,OAAA,CAAQ;AAAA,GACjB,CAAA;AACH;AA8BO,SAAS,gBAAA,CAAiB,OAAkB,MAAA,EAAwC;AACzF,EAAA,MAAM,QAAQ,KAAA,YAAiB/B,4BAAA,GAAa,KAAA,GAAQA,4BAAA,CAAW,WAAW,KAAK,CAAA;AAC/E,EAAA,MAAM,eAAe,MAAA,YAAkBA,4BAAA;AAEvC,EAAA,MAAM,KAAA,GACJ,MAAA,YAAkBA,4BAAA,GAAa,MAAA,GAASA,4BAAA,CAAW,WAAW,MAAA,EAAQ,EAAE,YAAA,EAAc,KAAA,EAAO,CAAA;AAE/F,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,MAAM,IAAIN,4BAAA,CAAW,CAAA,kDAAA,EAAqD,KAAA,CAAM,IAAI,CAAA,CAAE,CAAA;AAAA,EACxF;AAEA,EAAA,MAAM,QAAA,GAAW,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AACnC,EAAA,MAAM,QAAA,GAAW,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA;AAEnC,EAAA,IAAI,YAAA,GAAe,KAAA;AAGnB,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,IAAI,YAAA,EAAc;AAChB,MAAA,MAAM,IAAIA,6BAAW,0DAA0D,CAAA;AAAA,IACjF;AACA,IAAA,IAAI,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,KAAM,QAAA,EAAU;AAC/B,MAAA,MAAM,IAAIA,4BAAA;AAAA,QACR,yDAAyD,KAAA,CAAM,KAAA,CAAM,CAAC,CAAC,QAAQ,QAAQ,CAAA;AAAA,OACzF;AAAA,IACF;AAGA,IAAA,MAAM,MAAA,GAAS,QAAA,CAAS,KAAA,CAAM,MAAA,EAAQ,QAAQ,CAAA;AAC9C,IAAA,YAAA,GAAeM,6BAAW,UAAA,CAAW,MAAA,EAAQ,EAAE,YAAA,EAAc,OAAO,CAAA;AAAA,EACtE,CAAA,MAAA,IAAW,KAAA,CAAM,IAAA,KAAS,CAAA,EAAG;AAC3B,IAAA,IAAI,KAAA,CAAM,MAAM,CAAC,CAAA,KAAM,YAAY,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,KAAM,QAAA,EAAU;AAC9D,MAAA,MAAM,IAAIN,4BAAA;AAAA,QACR;AAAA,OACF;AAAA,IACF;AAAA,EACF,CAAA,MAAO;AACL,IAAA,MAAM,IAAIA,4BAAA,CAAW,CAAA,+CAAA,EAAkD,KAAA,CAAM,IAAI,CAAA,CAAA,CAAG,CAAA;AAAA,EACtF;AAGA,EAAA,MAAM,QAAA,GAAWiB,6BAAAA,CAAe,KAAA,EAAO,CAAC,CAAA;AAIxC,EAAA,MAAM,QAAA,GAAW,QAAA,CAAS,GAAA,CAAI,YAAY,CAAA;AAG1C,EAAA,MAAM,UAAA,GAAa,QAAA,CAAS,GAAA,CAAI,CAAC,CAAA;AASjC,EAAA,MAAM,QAAA,GAAW,UAAA,CAAW,IAAA,EAAK,CAAE,GAAA,EAAI;AACvC,EAAA,IAAI,EAAE,KAAA,YAAiBX,4BAAA,CAAA,IAAe,CAAC,YAAA,EAAc;AACnD,IAAA,MAAM,IAAA,GAAO,SAAS,MAAA,CAAO,IAAA;AAC7B,IAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,MAAA,MAAM,IAAIJ,6BAAW,gDAAgD,CAAA;AAAA,IACvE;AACA,IAAA,IAAI,gBAAgB,aAAA,EAAe;AACjC,MAAA,MAAM,GAAA,GAAMiC,kCAAA,CAAiB,IAAA,EAAM,QAAA,CAAS,OAAO,MAAM,CAAA;AACzD,MAAA,OAAO,OAAO,GAAG,CAAA;AAAA,IACnB;AACA,IAAA,OAAOC,mCAAA,CAAkB,IAAA,EAAM,QAAA,CAAS,MAAA,CAAO,MAAM,CAAA;AAAA,EACvD;AACA,EAAA,OAAO,QAAA;AACT;AAaO,SAAS,gCAAA,CACd,OACA,MAAA,EACqB;AACrB,EAAA,MAAM,QAAQ,KAAA,YAAiB9B,4BAAA,GAAa,KAAA,GAAQA,4BAAA,CAAW,WAAW,KAAK,CAAA;AAC/E,EAAA,MAAM,KAAA,GACJ,MAAA,YAAkBA,4BAAA,GAAa,MAAA,GAASA,4BAAA,CAAW,WAAW,MAAA,EAAQ,EAAE,YAAA,EAAc,KAAA,EAAO,CAAA;AAI/F,EAAA,IAAI,IAAA,GAAO,KAAA;AACX,EAAA,IAAI,KAAA,GAAQ,KAAA;AAEZ,EAAA,IAAI,IAAA,CAAK,IAAA,KAAS,CAAA,IAAK,IAAA,CAAK,SAAS,CAAA,EAAG;AACtC,IAAA,MAAM,IAAIN,6BAAW,kCAAkC,CAAA;AAAA,EACzD;AACA,EAAA,IAAI,KAAA,CAAM,IAAA,KAAS,CAAA,IAAK,KAAA,CAAM,SAAS,CAAA,EAAG;AACxC,IAAA,MAAM,IAAIA,6BAAW,mCAAmC,CAAA;AAAA,EAC1D;AAEA,EAAA,IAAI,IAAA,CAAK,SAAS,CAAA,EAAG;AACnB,IAAA,IAAA,GAAO,IAAA,CAAK,QAAQ,CAAC,IAAA,CAAK,MAAM,CAAC,CAAA,IAAK,CAAA,EAAG,CAAC,CAAC,CAAA;AAAA,EAC7C;AACA,EAAA,IAAI,KAAA,CAAM,SAAS,CAAA,EAAG;AACpB,IAAA,KAAA,GAAQ,KAAA,CAAM,QAAQ,CAAC,KAAA,CAAM,MAAM,CAAC,CAAA,IAAK,CAAA,EAAG,CAAC,CAAC,CAAA;AAAA,EAChD;AAEA,EAAA,IAAI,KAAK,IAAA,KAAS,CAAA,IAAK,KAAK,KAAA,CAAM,CAAC,MAAM,CAAA,EAAG;AAC1C,IAAA,MAAM,IAAIA,6BAAW,CAAA,oCAAA,CAAsC,CAAA;AAAA,EAC7D;AACA,EAAA,IAAI,MAAM,IAAA,KAAS,CAAA,IAAK,MAAM,KAAA,CAAM,CAAC,MAAM,CAAA,EAAG;AAC5C,IAAA,MAAM,IAAIA,6BAAW,CAAA,iDAAA,CAAmD,CAAA;AAAA,EAC1E;AACA,EAAA,IAAA,CAAK,IAAA,CAAK,MAAM,CAAC,CAAA,IAAK,QAAQ,KAAA,CAAM,KAAA,CAAM,CAAC,CAAA,IAAK,CAAA,CAAA,EAAI;AAClD,IAAA,MAAM,IAAIA,6BAAW,CAAA,mBAAA,CAAqB,CAAA;AAAA,EAC5C;AAEA,EAAA,MAAM,YAAY,IAAA,CAAK,KAAA;AACvB,EAAA,IAAI,cAAc,QAAA,EAAU;AAC1B,IAAA,MAAM,IAAIE,6BAAW,oDAAoD,CAAA;AAAA,EAC3E;AAMA,EAAA,MAAM,KAAA,GAAQ,KAAK,IAAA,EAAK;AAGxB,EAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,GAAA,CAAI,KAAK,CAAA;AAuB5B,EAAA,MAAM,OAAA,GAAU,KAAK,GAAA,EAAI;AACzB,EAAA,MAAM,UAAU,IAAA,CAAK,IAAA,GAAO,GAAA,CAAI,OAAA,CAAQ,MAAM,CAAA;AAE9C,EAAA,MAAM,SAAA,GAAY,OAAA,CAAQ,GAAA,EAAI,CAAE,GAAA,EAAI;AACpC,EAAA,MAAM,cAAc,SAAA,CAAU,KAAA;AAC9B,EAAA,IAAI,gBAAgB,QAAA,EAAU;AAC5B,IAAA,MAAM,IAAIA,6BAAW,gEAAgE,CAAA;AAAA,EACvF;AACA,EAAA,MAAM,MAAMI,4BAAA,CAAW,MAAA,CAAO,GAAG,EAAE,KAAA,EAAO,aAAa,CAAA;AACvD,EAAA,MAAM,KAAA,GAAQ,GAAA,CAAI,GAAA,CAAI,SAAS,EAAE,GAAA,EAAI;AAErC,EAAA,MAAM,IAAA,GAAO,MAAM,GAAA,CAAI,KAAK,EAAE,GAAA,CAAI,KAAK,EAAE,IAAA,EAAK;AAC9C,EAAA,IAAI,EAAE,KAAA,YAAiBA,4BAAA,CAAA,IAAe,EAAE,kBAAkBA,4BAAA,CAAA,EAAa;AACrE,IAAA,MAAM,IAAA,GAAO,KAAK,MAAA,CAAO,IAAA;AACzB,IAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,MAAA,MAAM,IAAIJ,6BAAW,gEAAgE,CAAA;AAAA,IACvF;AACA,IAAA,IAAI,gBAAgB,aAAA,EAAe;AACjC,MAAA,MAAM,GAAA,GAAMiC,kCAAA,CAAiB,IAAA,EAAM,IAAA,CAAK,OAAO,MAAM,CAAA;AACrD,MAAA,OAAO,OAAO,GAAG,CAAA;AAAA,IACnB;AACA,IAAA,OAAOC,mCAAA,CAAkB,IAAA,EAAM,IAAA,CAAK,MAAA,CAAO,MAAM,CAAA;AAAA,EACnD;AACA,EAAA,OAAO,IAAA;AACT;;;ACvPA,SAASa,YAAAA,CAAY,GAAsB,CAAA,EAA+B;AACxE,EAAA,IAAI,CAAA,CAAE,MAAA,KAAW,CAAA,CAAE,MAAA,EAAQ,OAAO,KAAA;AAClC,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,CAAA,CAAE,QAAQ,CAAA,EAAA,EAAK;AACjC,IAAA,IAAA,CAAK,CAAA,CAAE,CAAC,CAAA,IAAK,CAAA,OAAQ,EAAE,CAAC,CAAA,IAAK,IAAI,OAAO,KAAA;AAAA,EAC1C;AACA,EAAA,OAAO,IAAA;AACT;AAEA,SAAS,eAAA,CAAgB,CAAA,EAAW,CAAA,EAAW,OAAA,EAAuB;AACpE,EAAA,IAAI,CAACA,YAAAA,CAAY,CAAA,CAAE,KAAA,EAAO,CAAA,CAAE,KAAK,CAAA,EAAG;AAClC,IAAA,MAAM,IAAIjD,4BAAA,CAAW,CAAA,kBAAA,EAAqB,OAAO,CAAA,GAAA,EAAM,EAAE,KAAK,CAAA,MAAA,EAAS,CAAA,CAAE,KAAK,CAAA,CAAA,CAAG,CAAA;AAAA,EACnF;AACF;AAEA,SAAS,aAAA,CAAc,GAAW,OAAA,EAAuB;AACvD,EAAA,IAAI,CAAA,CAAE,UAAU,QAAA,EAAU;AACxB,IAAA,MAAM,IAAIE,4BAAA,CAAW,CAAA,EAAG,OAAO,CAAA,8BAAA,CAAgC,CAAA;AAAA,EACjE;AACF;AAIA,SAAS,iBAAA,CAAkB,WAAoC,OAAA,EAAuB;AACpF,EAAA,IAAI,SAAA,KAAc,MAAA,IAAU,SAAA,KAAc,KAAA,IAAS,cAAc,MAAA,EAAQ;AACvE,IAAA,MAAM,IAAIG,uCAAA;AAAA,MACR,GAAG,OAAO,CAAA,2CAAA,CAAA;AAAA,MACV,WAAA;AAAA,MACA;AAAA,KACF;AAAA,EACF;AACF;AAEA,SAAS,eAAA,CACP,IAAA,EACA,IAAA,EACA,cAAA,EACA,SACA,MAAA,EACQ;AACR,EAAA,MAAM,UAAA,GAAaD,qCAAA,CAAoB,IAAA,EAAM,cAAA,EAAgB,SAAS,MAAM,CAAA;AAC5E,EAAA,OAAOyC,oCAAA,CAAmB,MAAM,UAAU,CAAA;AAC5C;AAqCO,SAAS,OAAA,CACd,WAAA,EACA,OAAA,EACA,SAAA,GAAqC,MAAA,EAC7B;AACR,EAAA,iBAAA,CAAkB,WAAW,SAAS,CAAA;AACtC,EAAA,aAAA,CAAc,aAAa,SAAS,CAAA;AACpC,EAAA,aAAA,CAAc,SAAS,SAAS,CAAA;AAChC,EAAA,eAAA,CAAgB,WAAA,EAAa,SAAS,SAAS,CAAA;AAE/C,EAAA,MAAM,IAAA,GAAOK,qBAAA,CAAI,WAAA,EAAa,OAAO,CAAA;AACrC,EAAA,MAAM,WAAA,GAAcC,qBAAA,CAAI,IAAA,EAAMC,wBAAA,CAAO,CAAA,EAAG,EAAE,KAAA,EAAO,IAAA,CAAK,KAAA,EAAO,MAAA,EAAQ,IAAA,CAAK,MAAA,EAAQ,CAAC,CAAA;AAEnF,EAAA,IAAI,cAAc,MAAA,EAAQ;AACxB,IAAA,OAAO,WAAA;AAAA,EACT;AACA,EAAA,IAAI,cAAc,KAAA,EAAO;AACvB,IAAA,OAAOJ,sBAAI,WAAW,CAAA;AAAA,EACxB;AACA,EAAA,OAAOT,uBAAK,WAAW,CAAA;AACzB;AA0BO,SAAS,OAAA,CACd,WAAA,EACA,OAAA,EACA,SAAA,GAAqC,MAAA,EAC7B;AACR,EAAA,iBAAA,CAAkB,WAAW,SAAS,CAAA;AACtC,EAAA,aAAA,CAAc,aAAa,SAAS,CAAA;AACpC,EAAA,aAAA,CAAc,SAAS,SAAS,CAAA;AAChC,EAAA,eAAA,CAAgB,WAAA,EAAa,SAAS,SAAS,CAAA;AAE/C,EAAA,MAAM,IAAA,GAAOW,qBAAA,CAAI,WAAA,EAAa,OAAO,CAAA;AACrC,EAAA,MAAM,OAAA,GAAUG,sBAAI,IAAI,CAAA;AAExB,EAAA,IAAI,cAAc,MAAA,EAAQ;AACxB,IAAA,OAAO,OAAA;AAAA,EACT;AACA,EAAA,IAAI,cAAc,KAAA,EAAO;AACvB,IAAA,OAAOL,sBAAI,OAAO,CAAA;AAAA,EACpB;AACA,EAAA,OAAOT,uBAAK,OAAO,CAAA;AACrB;AA2BO,SAAS,sBAAA,CACd,WAAA,EACA,OAAA,EACA,SAAA,GAAqC,MAAA,EAC7B;AACR,EAAA,iBAAA,CAAkB,WAAW,wBAAwB,CAAA;AACrD,EAAA,aAAA,CAAc,aAAa,wBAAwB,CAAA;AACnD,EAAA,aAAA,CAAc,SAAS,wBAAwB,CAAA;AAC/C,EAAA,eAAA,CAAgB,WAAA,EAAa,SAAS,wBAAwB,CAAA;AAE9D,EAAA,MAAM,OAAA,GAAU,IAAA;AAChB,EAAA,MAAM,WAAA,GAAce,sBAAA,CAAK,WAAA,EAAa,OAAA,EAAS,IAAI,OAAO,CAAA;AAE1D,EAAA,MAAM,OAAA,GAAUC,sBAAI,WAAW,CAAA;AAC/B,EAAA,MAAM,KAAA,GAAQC,qBAAA,CAAI,OAAA,EAAS,OAAO,CAAA;AAElC,EAAA,MAAM,GAAA,GAAMJ,yBAAO,CAAA,EAAG;AAAA,IACpB,KAAA,EAAO,WAAA,CAAY,KAAA,KAAU,SAAA,GAAY,SAAA,GAAY,SAAA;AAAA,IACrD,QAAQ,WAAA,CAAY;AAAA,GACrB,CAAA;AACD,EAAA,MAAM,eAAA,GAAkBF,qBAAA,CAAI,GAAA,EAAK,OAAO,CAAA;AACxC,EAAA,MAAM,YAAA,GAAeA,qBAAA,CAAI,GAAA,EAAK,WAAW,CAAA;AACzC,EAAA,MAAM,eAAA,GAAkBK,sBAAI,YAAY,CAAA;AACxC,EAAA,MAAM,KAAA,GAAQC,qBAAA,CAAI,eAAA,EAAiB,eAAe,CAAA;AAElD,EAAA,MAAM,IAAA,GAAOC,qBAAA,CAAIzB,qBAAA,CAAI,KAAA,EAAO,KAAK,CAAC,CAAA;AAElC,EAAA,IAAI,cAAc,MAAA,EAAQ;AACxB,IAAA,OAAO,IAAA;AAAA,EACT;AACA,EAAA,IAAI,cAAc,KAAA,EAAO;AACvB,IAAA,OAAOgB,sBAAI,IAAI,CAAA;AAAA,EACjB;AACA,EAAA,OAAOT,uBAAK,IAAI,CAAA;AAClB;AAqBO,SAAS,QAAA,CAAS,aAAqB,OAAA,EAAyB;AACrE,EAAA,aAAA,CAAc,aAAa,UAAU,CAAA;AACrC,EAAA,aAAA,CAAc,SAAS,UAAU,CAAA;AACjC,EAAA,eAAA,CAAgB,WAAA,EAAa,SAAS,UAAU,CAAA;AAEhD,EAAA,MAAM,GAAA,GAAM,OAAA,CAAQ,WAAA,EAAa,OAAA,EAAS,MAAM,CAAA;AAChD,EAAA,OAAOmB,uBAAK,GAAG,CAAA;AACjB;AA6BO,SAAS,UACd,WAAA,EACA,OAAA,EACA,KAAA,GAAQ,CAAA,EACR,YAAqC,MAAA,EAC7B;AACR,EAAA,iBAAA,CAAkB,WAAW,WAAW,CAAA;AACxC,EAAA,aAAA,CAAc,aAAa,WAAW,CAAA;AACtC,EAAA,aAAA,CAAc,SAAS,WAAW,CAAA;AAClC,EAAA,eAAA,CAAgB,WAAA,EAAa,SAAS,WAAW,CAAA;AAEjD,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,IAAK,SAAS,CAAA,EAAG;AACzC,IAAA,MAAM,IAAIrD,uCAAA,CAAsB,CAAA,4BAAA,EAA+B,KAAK,CAAA,CAAA,EAAI,SAAS,KAAK,CAAA;AAAA,EACxF;AAEA,EAAA,MAAM,IAAA,GAAO6C,qBAAA,CAAI,WAAA,EAAa,OAAO,CAAA;AACrC,EAAA,MAAM,OAAA,GAAUG,sBAAI,IAAI,CAAA;AAExB,EAAA,MAAM,UAAU,OAAA,CAAQ,IAAA;AACxB,EAAA,IAAI,KAAA,CAAM,OAAA,CAAQ,OAAO,CAAA,EAAG;AAC1B,IAAA,MAAM,IAAInD,6BAAW,yCAAyC,CAAA;AAAA,EAChE;AACA,EAAA,MAAM,KAAA,GAAQ,WAAA,CAAY,KAAA,KAAU,SAAA,GAAY,SAAA,GAAY,SAAA;AAC5D,EAAA,MAAM,QAAA,GAAW,KAAA,KAAU,SAAA,GAAY,IAAI,YAAA,CAAa,IAAA,CAAK,IAAI,CAAA,GAAI,IAAI,YAAA,CAAa,IAAA,CAAK,IAAI,CAAA;AAC/F,EAAA,MAAM,cAAA,GAAiBC,gCAAA,CAAe,OAAA,CAAQ,KAAK,CAAA;AACnD,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,CAAK,MAAM,CAAA,EAAA,EAAK;AAClC,IAAA,MAAM,MAAA,GAAS,gBAAgB,OAAA,EAAS,CAAA,EAAG,gBAAgB,OAAA,CAAQ,OAAA,EAAS,QAAQ,MAAM,CAAA;AAC1F,IAAA,IAAI,UAAU,KAAA,EAAO;AACnB,MAAA,QAAA,CAAS,CAAC,CAAA,GAAI,GAAA,GAAM,MAAA,GAAS,MAAA;AAAA,IAC/B,CAAA,MAAO;AACL,MAAA,QAAA,CAAS,CAAC,CAAA,GAAI,KAAA,IAAS,MAAA,GAAS,GAAA,GAAM,KAAA,CAAA;AAAA,IACxC;AAAA,EACF;AAEA,EAAA,MAAM,IAAA,GAAOkC,yBAAO,cAAA,CAAe;AAAA,IACjC,IAAA,EAAM,QAAA;AAAA,IACN,OAAO,WAAA,CAAY,KAAA;AAAA,IACnB,KAAA;AAAA,IACA,QAAQ,WAAA,CAAY;AAAA,GACrB,CAAA;AAED,EAAA,IAAI,cAAc,MAAA,EAAQ;AACxB,IAAA,OAAO,IAAA;AAAA,EACT;AACA,EAAA,IAAI,cAAc,KAAA,EAAO;AACvB,IAAA,OAAOW,sBAAI,IAAI,CAAA;AAAA,EACjB;AACA,EAAA,OAAOT,uBAAK,IAAI,CAAA;AAClB","file":"chunk-QERHVCHC.cjs","sourcesContent":["// Module base class for neural networks\n\n// Containers\nexport { Sequential } from \"./containers/Sequential\";\n// Activation layers\nexport {\n  ELU,\n  GELU,\n  LeakyReLU,\n  LogSoftmax,\n  Mish,\n  ReLU,\n  Sigmoid,\n  Softmax,\n  Softplus,\n  Swish,\n  Tanh,\n} from \"./layers/activations\";\n// Attention layers\nexport {\n  MultiheadAttention,\n  TransformerEncoderLayer,\n} from \"./layers/attention\";\n// Convolutional layers\nexport { AvgPool2d, Conv1d, Conv2d, MaxPool2d } from \"./layers/conv\";\n// Regularization layers\nexport { Dropout } from \"./layers/dropout\";\n// Layers - fully connected / dense layers\nexport { Linear } from \"./layers/linear\";\n// Normalization layers\nexport { BatchNorm1d, LayerNorm } from \"./layers/normalization\";\n// Recurrent layers\nexport { GRU, LSTM, RNN } from \"./layers/recurrent\";\n// Loss functions\nexport {\n  binaryCrossEntropyLoss,\n  binaryCrossEntropyWithLogitsLoss,\n  crossEntropyLoss,\n  huberLoss,\n  maeLoss,\n  mseLoss,\n  rmseLoss,\n} from \"./losses/index\";\nexport type { ForwardHook, ForwardPreHook } from \"./module/Module\";\nexport { Module } from \"./module/Module\";\n","import {\n  DeepboxError,\n  type Device,\n  type DType,\n  DTypeError,\n  InvalidParameterError,\n  isDevice,\n  ShapeError,\n} from \"../../core\";\nimport { type AnyTensor, GradTensor, type Tensor } from \"../../ndarray\";\nimport { offsetFromFlatIndex } from \"../../ndarray/tensor/strides\";\nimport { computeStrides } from \"../../ndarray/tensor/Tensor\";\n\ntype StateEntry = {\n  data: Array<number | string | bigint>;\n  dtype: DType;\n  shape: number[];\n};\n\nfunction shapesEqual(a: readonly number[], b: readonly number[]): boolean {\n  if (a.length !== b.length) return false;\n  for (let i = 0; i < a.length; i++) {\n    if ((a[i] ?? 0) !== (b[i] ?? 0)) return false;\n  }\n  return true;\n}\n\nfunction sizeFromShape(shape: readonly number[], context: string): number {\n  let size = 1;\n  for (const dim of shape) {\n    if (!Number.isInteger(dim) || dim < 0) {\n      throw new ShapeError(`${context} contains invalid dimension ${String(dim)}`);\n    }\n    size *= dim;\n  }\n  return size;\n}\n\nfunction cloneTensorData(t: Tensor): Array<number | string | bigint> {\n  const data = t.data;\n  if (Array.isArray(data)) {\n    return data.slice();\n  }\n  if (data instanceof BigInt64Array) {\n    return Array.from(data);\n  }\n  const out = new Array<number>(data.length);\n  for (let i = 0; i < data.length; i++) {\n    const value = data[i];\n    if (value === undefined) {\n      throw new DeepboxError(\"Internal error: tensor data access out of bounds\");\n    }\n    out[i] = value;\n  }\n  return out;\n}\n\nfunction validateStateEntryShape(\n  name: string,\n  kind: \"parameter\" | \"buffer\",\n  entry: StateEntry\n): void {\n  const size = sizeFromShape(entry.shape, `${kind} ${name} shape`);\n  if (entry.data.length !== size) {\n    throw new ShapeError(\n      `${kind} ${name} data length ${entry.data.length} does not match shape size ${size}`\n    );\n  }\n}\n\nfunction copyStateEntryIntoTensor(\n  name: string,\n  kind: \"parameter\" | \"buffer\",\n  target: Tensor,\n  entry: StateEntry\n): void {\n  if (!shapesEqual(target.shape, entry.shape)) {\n    throw new ShapeError(\n      `${kind} ${name} shape mismatch: expected [${target.shape.join(\", \")}], got [${entry.shape.join(\", \")}]`\n    );\n  }\n  if (target.dtype !== entry.dtype) {\n    throw new DTypeError(\n      `${kind} ${name} dtype mismatch: expected ${target.dtype}, got ${entry.dtype}`\n    );\n  }\n\n  const size = sizeFromShape(entry.shape, `${kind} ${name} shape`);\n  const logicalStrides = computeStrides(target.shape);\n  const data = target.data;\n\n  if (target.dtype === \"string\") {\n    if (!Array.isArray(data)) {\n      throw new DTypeError(`${kind} ${name} expected string data`);\n    }\n    for (let i = 0; i < size; i++) {\n      const value = entry.data[i];\n      if (typeof value !== \"string\") {\n        throw new DTypeError(`${kind} ${name} expects string data`);\n      }\n      const offset = offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);\n      data[offset] = value;\n    }\n    return;\n  }\n\n  if (data instanceof BigInt64Array) {\n    for (let i = 0; i < size; i++) {\n      const value = entry.data[i];\n      if (typeof value !== \"bigint\") {\n        throw new DTypeError(`${kind} ${name} expects bigint data`);\n      }\n      const offset = offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);\n      data[offset] = value;\n    }\n    return;\n  }\n\n  if (Array.isArray(data)) {\n    throw new DTypeError(`${kind} ${name} expected numeric data`);\n  }\n\n  for (let i = 0; i < size; i++) {\n    const value = entry.data[i];\n    if (typeof value !== \"number\") {\n      throw new DTypeError(`${kind} ${name} expects numeric data`);\n    }\n    const offset = offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);\n    data[offset] = value;\n  }\n}\n\n/**\n * Hook function called before the forward pass.\n *\n * @param module - The module being called\n * @param inputs - The input tensors to the forward pass\n * @returns Modified inputs array, or undefined to keep original inputs\n */\nexport type ForwardPreHook = (module: Module, inputs: AnyTensor[]) => AnyTensor[] | undefined;\n\n/**\n * Hook function called after the forward pass.\n *\n * @param module - The module being called\n * @param inputs - The input tensors to the forward pass\n * @param output - The output tensor from the forward pass\n * @returns Modified output tensor, or undefined to keep original output\n */\nexport type ForwardHook = (\n  module: Module,\n  inputs: AnyTensor[],\n  output: AnyTensor\n) => AnyTensor | undefined;\n\n/**\n * Base class for all neural network modules.\n *\n * All models should subclass this class. Modules can contain other modules,\n * allowing to nest them in a tree structure.\n *\n * This is analogous to PyTorch's nn.Module.\n *\n * @example\n * ```ts\n * import { Module, Linear, ReLU } from 'deepbox/nn';\n * import type { Tensor } from 'deepbox/ndarray';\n *\n * class MyModel extends Module {\n *   private fc1: Linear;\n *   private relu: ReLU;\n *   private fc2: Linear;\n *\n *   constructor() {\n *     super();\n *     this.fc1 = new Linear(10, 5);\n *     this.relu = new ReLU();\n *     this.fc2 = new Linear(5, 2);\n *     this.registerModule('fc1', this.fc1);\n *     this.registerModule('relu', this.relu);\n *     this.registerModule('fc2', this.fc2);\n *   }\n *\n *   forward(x: Tensor): Tensor {\n *     let out = this.fc1.forward(x);\n *     out = this.relu.forward(out);\n *     out = this.fc2.forward(out);\n *     return out;\n *   }\n * }\n * ```\n *\n * References:\n * - PyTorch nn.Module: https://pytorch.org/docs/stable/generated/torch.nn.Module.html\n *\n * @category Neural Networks\n */\nexport abstract class Module {\n  /** Child modules registered to this module - stores nested layers/modules */\n  protected _modules: Map<string, Module> = new Map();\n\n  /** Parameters of this module - trainable tensors (weights, biases) wrapped as GradTensor */\n  protected _parameters: Map<string, GradTensor> = new Map();\n\n  /** Buffers (non-trainable tensors) of this module - e.g., running stats in BatchNorm */\n  protected _buffers: Map<string, Tensor> = new Map();\n\n  /** Training mode flag - affects behavior of layers like Dropout and BatchNorm */\n  protected _training = true;\n\n  /** Forward pre-hooks registered on this module */\n  private _forwardPreHooks: Map<number, ForwardPreHook> = new Map();\n  /** Forward hooks registered on this module */\n  private _forwardHooks: Map<number, ForwardHook> = new Map();\n  /** Incrementing hook id */\n  private _nextHookId = 0;\n\n  /**\n   * Forward pass of the module.\n   *\n   * Should be overridden by all subclasses. Accepts either regular Tensors\n   * or GradTensors for automatic differentiation support.\n   *\n   * @param inputs - Input tensors (Tensor or GradTensor)\n   * @returns Output tensor (Tensor or GradTensor depending on input and layer type)\n   *\n   * @example\n   * ```ts\n   * // Using with regular Tensor\n   * const output = model.forward(inputTensor);\n   *\n   * // Using with GradTensor for training\n   * const gradOutput = model.forward(gradInput);\n   * gradOutput.backward();\n   * ```\n   */\n  abstract forward(...inputs: AnyTensor[]): AnyTensor;\n\n  /**\n   * Makes the module callable (allows using `module(x)` instead of `module.forward(x)`).\n   *\n   * @param inputs - Input tensors (Tensor or GradTensor)\n   * @returns Output tensor\n   */\n  call(...inputs: AnyTensor[]): AnyTensor {\n    let curInputs = inputs;\n    for (const hook of this._forwardPreHooks.values()) {\n      const result = hook(this, curInputs);\n      if (Array.isArray(result)) {\n        curInputs = result;\n      }\n    }\n    let output = this.forward(...curInputs);\n    for (const hook of this._forwardHooks.values()) {\n      const result = hook(this, curInputs, output);\n      if (result !== undefined) {\n        output = result;\n      }\n    }\n    return output;\n  }\n\n  /**\n   * Register a child module.\n   *\n   * @param name - Name of the module\n   * @param module - The module to register\n   */\n  protected registerModule(name: string, module: Module): void {\n    // Store the child module in the modules map for hierarchical tracking\n    this._modules.set(name, module);\n  }\n\n  /**\n   * Register a parameter (trainable tensor).\n   *\n   * Parameters must be GradTensor instances with requiresGrad=true for\n   * proper gradient computation during backpropagation.\n   *\n   * @param name - Name of the parameter\n   * @param param - The parameter tensor (must be GradTensor)\n   */\n  protected registerParameter(name: string, param: GradTensor): void {\n    // Register a trainable parameter (weight or bias) for optimization\n    this._parameters.set(name, param);\n  }\n\n  /**\n   * Register a buffer (non-trainable tensor).\n   *\n   * Buffers are typically used for running statistics in batch normalization.\n   *\n   * @param name - Name of the buffer\n   * @param buffer - The buffer tensor\n   */\n  protected registerBuffer(name: string, buffer: Tensor): void {\n    // Register a non-trainable buffer (e.g., running mean/variance in BatchNorm)\n    // Buffers are saved with the model but not updated by optimizers\n    this._buffers.set(name, buffer);\n  }\n\n  /**\n   * Get all parameters of this module and its children.\n   *\n   * Returns GradTensor instances that are compatible with optimizers.\n   * This enables direct usage with optimizer constructors:\n   * ```ts\n   * const optimizer = new Adam(model.parameters());\n   * ```\n   *\n   * @param recurse - Whether to include parameters of child modules\n   * @returns Iterator of GradTensor parameters\n   */\n  *parameters(recurse = true): Generator<GradTensor> {\n    // Yield own parameters first\n    for (const param of this._parameters.values()) {\n      yield param;\n    }\n\n    // Recursively yield child module parameters if requested\n    // This allows optimizers to access all trainable parameters in the model\n    if (recurse) {\n      for (const module of this._modules.values()) {\n        yield* module.parameters(true);\n      }\n    }\n  }\n\n  /**\n   * Get all named parameters of this module and its children.\n   *\n   * @param prefix - Prefix for parameter names\n   * @param recurse - Whether to include parameters of child modules\n   * @returns Iterator of [name, parameter] pairs\n   */\n  *namedParameters(prefix = \"\", recurse = true): Generator<[string, GradTensor]> {\n    // Yield own parameters with hierarchical naming (e.g., \"fc1.weight\")\n    for (const [name, param] of this._parameters.entries()) {\n      // Build full parameter name with dot notation for nested modules\n      const fullName = prefix ? `${prefix}.${name}` : name;\n      yield [fullName, param];\n    }\n\n    // Recursively yield child module parameters with proper prefixing\n    if (recurse) {\n      for (const [moduleName, module] of this._modules.entries()) {\n        // Extend prefix for nested modules (e.g., \"encoder.fc1\")\n        const fullPrefix = prefix ? `${prefix}.${moduleName}` : moduleName;\n        yield* module.namedParameters(fullPrefix, true);\n      }\n    }\n  }\n\n  /**\n   * Get all child modules.\n   *\n   * @param recurse - Whether to include nested child modules\n   * @returns Iterator of modules\n   */\n  *modules(recurse = true): Generator<Module> {\n    // Always yield self first (root of the module tree)\n    yield this;\n\n    // Recursively yield all child modules in depth-first order\n    if (recurse) {\n      for (const module of this._modules.values()) {\n        yield* module.modules(true);\n      }\n    }\n  }\n\n  /**\n   * Get all named child modules.\n   *\n   * @param prefix - Prefix for module names\n   * @param recurse - Whether to include nested child modules\n   * @returns Iterator of [name, module] pairs\n   */\n  *namedModules(prefix = \"\", recurse = true): Generator<[string, Module]> {\n    // Yield self with current prefix (empty string for root)\n    yield [prefix, this];\n\n    // Recursively yield child modules with hierarchical naming\n    if (recurse) {\n      for (const [name, module] of this._modules.entries()) {\n        // Build full module path (e.g., \"encoder.layer1\")\n        const fullName = prefix ? `${prefix}.${name}` : name;\n        yield* module.namedModules(fullName, true);\n      }\n    }\n  }\n\n  /**\n   * Set the module in training mode.\n   *\n   * This affects certain layers like Dropout and BatchNorm.\n   *\n   * @param mode - Training mode (true) or evaluation mode (false)\n   * @returns this\n   */\n  train(mode = true): this {\n    // Set training mode for this module\n    this._training = mode;\n\n    // Recursively propagate training mode to all child modules\n    // This ensures layers like Dropout and BatchNorm behave correctly\n    for (const module of this._modules.values()) {\n      module.train(mode);\n    }\n\n    // Return this for method chaining (e.g., model.train().forward(x))\n    return this;\n  }\n\n  /**\n   * Set the module in evaluation mode.\n   *\n   * This is equivalent to calling `train(false)`.\n   *\n   * @returns this\n   */\n  eval(): this {\n    return this.train(false);\n  }\n\n  /**\n   * Check if the module is in training mode.\n   *\n   * @returns true if in training mode\n   */\n  get training(): boolean {\n    return this._training;\n  }\n\n  /**\n   * Zero out the gradients of all parameters.\n   *\n   * Call this before each training iteration to prevent gradient accumulation\n   * from previous iterations.\n   *\n   * For parameters wrapped in GradTensor, this calls zeroGrad() on each.\n   * For regular Tensors, this is a no-op until they are converted to GradTensor.\n   *\n   * @example\n   * ```ts\n   * model.zeroGrad();\n   * const output = model.forward(input);\n   * // ... compute loss and backward\n   * optimizer.step();\n   * ```\n   */\n  zeroGrad(): void {\n    // Zero out gradients for all parameters in the module\n    // This should be called before each backward pass to prevent gradient accumulation\n    for (const param of this.parameters()) {\n      // parameters() yields GradTensor instances, so zeroGrad is always available\n      param.zeroGrad();\n    }\n  }\n\n  /**\n   * Get all buffers of this module and its children.\n   */\n  *buffers(recurse = true): Generator<Tensor> {\n    for (const buffer of this._buffers.values()) {\n      yield buffer;\n    }\n    if (recurse) {\n      for (const module of this._modules.values()) {\n        yield* module.buffers(true);\n      }\n    }\n  }\n\n  /**\n   * Get all named buffers of this module and its children.\n   */\n  *namedBuffers(prefix = \"\", recurse = true): Generator<[string, Tensor]> {\n    for (const [name, buffer] of this._buffers.entries()) {\n      const fullName = prefix ? `${prefix}.${name}` : name;\n      yield [fullName, buffer];\n    }\n    if (recurse) {\n      for (const [moduleName, module] of this._modules.entries()) {\n        const fullPrefix = prefix ? `${prefix}.${moduleName}` : moduleName;\n        yield* module.namedBuffers(fullPrefix, true);\n      }\n    }\n  }\n\n  /**\n   * Freeze specific parameters by name (or all if none provided).\n   *\n   * **⚠️ IMPORTANT**: This method creates new GradTensor instances with updated\n   * `requiresGrad` flags. Any external references to the old parameter objects\n   * will become stale. If you're using an optimizer that holds parameter references,\n   * you should recreate the optimizer after freezing/unfreezing parameters.\n   *\n   * @param names - Array of parameter names to freeze (e.g., ['fc1.weight']). If undefined, freezes all parameters.\n   * @param recurse - Whether to include parameters from child modules (default: true)\n   *\n   * @example\n   * ```ts\n   * const model = new MyModel();\n   * // Freeze only the first layer's weights\n   * model.freezeParameters(['fc1.weight']);\n   * // Note: Recreate optimizer after freezing\n   * const optimizer = new Adam(model.parameters());\n   * ```\n   */\n  freezeParameters(names?: string[], recurse = true): void {\n    this.setRequiresGradForNames(names, false, recurse);\n  }\n\n  /**\n   * Unfreeze specific parameters by name (or all if none provided).\n   *\n   * **⚠️ IMPORTANT**: This method creates new GradTensor instances with updated\n   * `requiresGrad` flags. Any external references to the old parameter objects\n   * will become stale. If you're using an optimizer that holds parameter references,\n   * you should recreate the optimizer after freezing/unfreezing parameters.\n   *\n   * @param names - Array of parameter names to unfreeze (e.g., ['fc1.weight']). If undefined, unfreezes all parameters.\n   * @param recurse - Whether to include parameters from child modules (default: true)\n   *\n   * @example\n   * ```ts\n   * const model = new MyModel();\n   * model.freezeParameters(); // Freeze all\n   * model.unfreezeParameters(['fc2.weight']); // Unfreeze only fc2 weights\n   * // Note: Recreate optimizer after unfreezing\n   * const optimizer = new Adam(model.parameters());\n   * ```\n   */\n  unfreezeParameters(names?: string[], recurse = true): void {\n    this.setRequiresGradForNames(names, true, recurse);\n  }\n\n  private setRequiresGradForNames(\n    names: string[] | undefined,\n    requiresGrad: boolean,\n    recurse: boolean\n  ): void {\n    const providedNames = names !== undefined;\n    const targetNames =\n      names ?? Array.from(this.namedParameters(\"\", recurse)).map(([name]) => name);\n    for (const name of targetNames) {\n      const resolved = this.resolveModuleAndName(name);\n      if (!resolved) {\n        if (providedNames) {\n          throw new InvalidParameterError(`Unknown parameter name: ${name}`, \"names\", name);\n        }\n        continue;\n      }\n      const { module, localName } = resolved;\n      const param = module._parameters.get(localName);\n      if (!param) {\n        if (providedNames) {\n          throw new InvalidParameterError(`Unknown parameter name: ${name}`, \"names\", name);\n        }\n        continue;\n      }\n      // Replace parameter to ensure requiresGrad flag change is reflected consistently.\n      const nextParam = GradTensor.fromTensor(param.tensor, { requiresGrad });\n      module._parameters.set(localName, nextParam);\n      for (const [key, value] of Object.entries(module)) {\n        if (value === param) {\n          Reflect.set(module, key, nextParam);\n        }\n      }\n    }\n  }\n\n  private resolveModuleAndName(fullName: string): { module: Module; localName: string } | null {\n    const parts = fullName.split(\".\");\n    let module: Module = this;\n    for (let i = 0; i < parts.length - 1; i++) {\n      const part = parts[i] ?? \"\";\n      const child = module._modules.get(part);\n      if (!child) return null;\n      module = child;\n    }\n    const localName = parts[parts.length - 1] ?? \"\";\n    return { module, localName };\n  }\n\n  private static setTensorDeviceMetadata(target: Tensor, device: Device): void {\n    if (!Reflect.set(target, \"device\", device)) {\n      throw new DeepboxError(\"Failed to update tensor device metadata\");\n    }\n  }\n\n  /**\n   * Get the state dictionary of the module.\n   */\n  stateDict(): {\n    parameters: Record<string, StateEntry>;\n    buffers: Record<string, StateEntry>;\n  } {\n    const parameters: Record<string, StateEntry> = {};\n    const buffers: Record<string, StateEntry> = {};\n\n    for (const [name, param] of this.namedParameters()) {\n      const t = param.tensor;\n      const data = cloneTensorData(t);\n      parameters[name] = {\n        data,\n        shape: [...t.shape],\n        dtype: t.dtype,\n      };\n    }\n\n    for (const [name, buffer] of this.namedBuffers()) {\n      const data = cloneTensorData(buffer);\n      buffers[name] = {\n        data,\n        shape: [...buffer.shape],\n        dtype: buffer.dtype,\n      };\n    }\n\n    return { parameters, buffers };\n  }\n\n  /**\n   * Load state dictionary into the module.\n   */\n  loadStateDict(stateDict: {\n    parameters?: Record<string, StateEntry>;\n    buffers?: Record<string, StateEntry>;\n  }): void {\n    const parameters = stateDict.parameters ?? {};\n    const buffers = stateDict.buffers ?? {};\n\n    const namedParams = new Map(this.namedParameters());\n    const namedBuffs = new Map(this.namedBuffers());\n\n    for (const name of namedParams.keys()) {\n      if (!(name in parameters)) {\n        throw new InvalidParameterError(`missing parameter: ${name}`, \"stateDict.parameters\", name);\n      }\n    }\n\n    for (const name of namedBuffs.keys()) {\n      if (!(name in buffers)) {\n        throw new InvalidParameterError(`missing buffer: ${name}`, \"stateDict.buffers\", name);\n      }\n    }\n\n    for (const name of Object.keys(parameters)) {\n      if (!namedParams.has(name)) {\n        throw new InvalidParameterError(\n          `unexpected parameter: ${name}`,\n          \"stateDict.parameters\",\n          name\n        );\n      }\n    }\n\n    for (const name of Object.keys(buffers)) {\n      if (!namedBuffs.has(name)) {\n        throw new InvalidParameterError(`unexpected buffer: ${name}`, \"stateDict.buffers\", name);\n      }\n    }\n\n    for (const [name, entry] of Object.entries(parameters)) {\n      const param = namedParams.get(name);\n      if (!param) continue;\n      validateStateEntryShape(name, \"parameter\", entry);\n      copyStateEntryIntoTensor(name, \"parameter\", param.tensor, entry);\n    }\n\n    for (const [name, entry] of Object.entries(buffers)) {\n      const buffer = namedBuffs.get(name);\n      if (!buffer) continue;\n      validateStateEntryShape(name, \"buffer\", entry);\n      copyStateEntryIntoTensor(name, \"buffer\", buffer, entry);\n    }\n  }\n\n  /**\n   * Move module to a specific device.\n   *\n   * **⚠️ WARNING**: This is a metadata-only operation. It updates the device\n   * property on parameters and buffers but does NOT actually transfer data\n   * between devices. Actual device data transfer requires device-specific\n   * memory management which is not yet implemented.\n   *\n   * This method is provided for API compatibility and future extensibility.\n   * Currently, it only updates the `device` metadata field.\n   *\n   * @param device - Target device identifier (e.g., 'cpu', 'webgpu', 'wasm')\n   * @returns this module for method chaining\n   *\n   * @example\n   * ```ts\n   * const model = new Linear(10, 5);\n   * model.to('webgpu'); // Updates device metadata only\n   * ```\n   */\n  to(device: Device): this {\n    if (!isDevice(device)) {\n      throw new InvalidParameterError(\"device must be one of: cpu, webgpu, wasm\", \"device\", device);\n    }\n\n    for (const param of this.parameters()) {\n      Module.setTensorDeviceMetadata(param.tensor, device);\n    }\n    for (const buffer of this.buffers()) {\n      Module.setTensorDeviceMetadata(buffer, device);\n    }\n    return this;\n  }\n\n  /**\n   * Apply a function to all modules recursively.\n   */\n  apply(fn: (module: Module) => void): this {\n    for (const module of this.modules()) {\n      fn(module);\n    }\n    return this;\n  }\n\n  /**\n   * Register a forward pre-hook.\n   */\n  registerForwardPreHook(hook: ForwardPreHook): () => void {\n    const hookId = this._nextHookId++;\n    this._forwardPreHooks.set(hookId, hook);\n    return () => {\n      this._forwardPreHooks.delete(hookId);\n    };\n  }\n\n  /**\n   * Register a forward hook.\n   */\n  registerForwardHook(hook: ForwardHook): () => void {\n    const hookId = this._nextHookId++;\n    this._forwardHooks.set(hookId, hook);\n    return () => {\n      this._forwardHooks.delete(hookId);\n    };\n  }\n\n  /**\n   * Get string representation of the module.\n   *\n   * @returns Hierarchical string representation showing module structure\n   */\n  toString(): string {\n    const lines = [`${this.constructor.name}(`];\n\n    // Iterate through child modules and format them with indentation\n    for (const [name, module] of this._modules.entries()) {\n      // Recursively get child module's string representation\n      const childLines = module.toString().split(\"\\n\");\n      // First line goes on the same line as the name; subsequent lines are indented\n      const moduleStr = childLines.map((line, i) => (i === 0 ? line : `  ${line}`)).join(\"\\n\");\n      // Format as: (name): ModuleType(...)\n      lines.push(`  (${name}): ${moduleStr}`);\n    }\n\n    lines.push(\")\");\n    return lines.join(\"\\n\");\n  }\n}\n","import { DeepboxError, IndexError, InvalidParameterError } from \"../../core\";\nimport type { AnyTensor } from \"../../ndarray\";\nimport { Module } from \"../module/Module\";\n\n/**\n * Sequential container for stacking layers in a linear pipeline.\n *\n * **Purpose:**\n * - Simplifies model construction by chaining layers sequentially\n * - Automatically manages forward pass through all layers\n * - Provides clean API for building feedforward networks\n *\n * **Behavior:**\n * The output of each layer becomes the input to the next layer.\n * Layers are executed in the order they were added.\n *\n * @example\n * ```ts\n * import { Sequential, Linear, ReLU, Dropout } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * // Create a simple feedforward network\n * const model = new Sequential(\n *   new Linear(784, 256),\n *   new ReLU(),\n *   new Dropout(0.5),\n *   new Linear(256, 10)\n * );\n *\n * const input = tensor(new Array(784).fill(0));\n * const output = model.forward(input);\n * ```\n *\n * @example\n * ```ts\n * // Access individual layers\n * const model = new Sequential(\n *   new Linear(10, 5),\n *   new ReLU()\n * );\n *\n * const firstLayer = model.getLayer(0); // Linear layer\n * const layerCount = model.length; // 2\n * ```\n *\n * References:\n * - PyTorch Sequential: https://pytorch.org/docs/stable/generated/torch.nn.Sequential.html\n * - Keras Sequential: https://keras.io/guides/sequential_model/\n *\n * @category Neural Network Containers\n */\nexport class Sequential extends Module {\n  /** Array of layers in sequential order */\n  private readonly layers: Module[];\n\n  /**\n   * Create a new Sequential container.\n   *\n   * @param layers - Variable number of Module instances to stack sequentially\n   * @throws {InvalidParameterError} If no layers are provided\n   * @throws {DeepboxError} If a layer is undefined\n   */\n  constructor(...layers: Module[]) {\n    super();\n\n    // Validate that at least one layer is provided\n    if (layers.length === 0) {\n      throw new InvalidParameterError(\n        \"Sequential requires at least one layer\",\n        \"layers\",\n        layers.length\n      );\n    }\n\n    // Store layers in execution order\n    this.layers = layers;\n\n    // Register each layer as a child module with numeric index as name\n    // This enables parameter tracking and hierarchical naming\n    for (let i = 0; i < layers.length; i++) {\n      const layer = layers[i];\n      if (!layer) {\n        throw new DeepboxError(`Layer at index ${i} is undefined`);\n      }\n      this.registerModule(String(i), layer);\n    }\n  }\n\n  /**\n   * Forward pass: sequentially apply all layers.\n   *\n   * The output of each layer becomes the input to the next layer.\n   *\n   * @param input - Input tensor (Tensor or GradTensor)\n   * @returns Output tensor after passing through all layers\n   * @throws {InvalidParameterError} If the input count is invalid or a layer returns multiple outputs\n   * @throws {DeepboxError} If a layer is undefined\n   */\n  forward(...inputs: AnyTensor[]): AnyTensor {\n    if (inputs.length !== 1) {\n      throw new InvalidParameterError(\n        \"Sequential.forward expects a single input tensor\",\n        \"inputs\",\n        inputs.length\n      );\n    }\n    const input = inputs[0];\n    if (!input) {\n      throw new InvalidParameterError(\n        \"Sequential.forward expects a single input tensor\",\n        \"input\",\n        input\n      );\n    }\n    // Start with the input tensor\n    let output = input;\n\n    // Sequentially apply each layer's forward pass\n    // Each layer transforms the output from the previous layer\n    for (let i = 0; i < this.layers.length; i++) {\n      const layer = this.layers[i];\n      if (!layer) {\n        throw new DeepboxError(`Layer at index ${i} is undefined`);\n      }\n\n      // Apply current layer's transformation\n      // Type assertion needed because forward can return Tensor | Tensor[]\n      const result = layer.call(output);\n      if (Array.isArray(result)) {\n        throw new InvalidParameterError(\n          `Sequential does not support layers that return multiple tensors (layer ${i})`,\n          \"layer\",\n          i\n        );\n      }\n      output = result;\n    }\n\n    // Return final output after all transformations\n    return output;\n  }\n\n  /**\n   * Get a layer by index.\n   *\n   * @param index - Zero-based index of the layer\n   * @returns The layer at the specified index\n   * @throws {IndexError} If index is out of bounds\n   * @throws {DeepboxError} If a layer is undefined\n   */\n  getLayer(index: number): Module {\n    // Validate index is within bounds\n    if (index < 0 || index >= this.layers.length) {\n      throw new IndexError(`Layer index ${index} out of bounds [0, ${this.layers.length})`, {\n        index,\n        validRange: [0, this.layers.length - 1],\n      });\n    }\n\n    const layer = this.layers[index];\n    if (!layer) {\n      throw new DeepboxError(`Layer at index ${index} is undefined`);\n    }\n\n    return layer;\n  }\n\n  /**\n   * Get the number of layers in the sequential container.\n   */\n  get length(): number {\n    return this.layers.length;\n  }\n\n  /**\n   * Get string representation showing all layers.\n   *\n   * @returns Multi-line string with each layer on a separate line\n   */\n  override toString(): string {\n    // Build hierarchical representation\n    const lines = [\"Sequential(\"];\n\n    // Add each layer with its index\n    for (let i = 0; i < this.layers.length; i++) {\n      const layer = this.layers[i];\n      if (!layer) continue;\n\n      // Get layer's string representation and indent continuation lines\n      const childLines = layer.toString().split(\"\\n\");\n      const layerStr = childLines.map((line, idx) => (idx === 0 ? line : `  ${line}`)).join(\"\\n\");\n\n      // Format as: (index): LayerType(...)\n      lines.push(`  (${i}): ${layerStr}`);\n    }\n\n    lines.push(\")\");\n    return lines.join(\"\\n\");\n  }\n\n  /**\n   * Iterate over all layers.\n   *\n   * @returns Iterator of layers\n   */\n  *[Symbol.iterator](): Iterator<Module> {\n    for (const layer of this.layers) {\n      yield layer;\n    }\n  }\n}\n","import { type Axis, normalizeAxis } from \"../../core\";\nimport { GradTensor, type Tensor } from \"../../ndarray\";\nimport { logSoftmax as gradLogSoftmax, softmax as gradSoftmax } from \"../../ndarray/autograd\";\nimport {\n  elu as eluOp,\n  gelu as geluOp,\n  leakyRelu as leakyReluOp,\n  logSoftmax as logSoftmaxOp,\n  mish as mishOp,\n  relu as reluOp,\n  sigmoid as sigmoidOp,\n  softmax as softmaxOp,\n  softplus as softplusOp,\n  swish as swishOp,\n} from \"../../ndarray/ops/activation\";\nimport { tanh as tanhOp } from \"../../ndarray/ops/trigonometry\";\nimport { Module } from \"../module/Module\";\n\n/**\n * Applies the Rectified Linear Unit (ReLU) activation function element-wise.\n *\n * ReLU(x) = max(0, x)\n *\n * @category Neural Network Layers\n */\nexport class ReLU extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) return input.relu();\n    return reluOp(input);\n  }\n\n  override toString(): string {\n    return \"ReLU()\";\n  }\n}\n\n/**\n * Applies the Sigmoid activation function element-wise.\n *\n * Sigmoid(x) = 1 / (1 + exp(-x))\n *\n * @category Neural Network Layers\n */\nexport class Sigmoid extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) return input.sigmoid();\n    return sigmoidOp(input);\n  }\n\n  override toString(): string {\n    return \"Sigmoid()\";\n  }\n}\n\n/**\n * Applies the Hyperbolic Tangent (Tanh) activation function element-wise.\n *\n * Tanh(x) = (exp(x) - exp(-x)) / (exp(x) + exp(-x))\n *\n * @category Neural Network Layers\n */\nexport class Tanh extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) return input.tanh();\n    return tanhOp(input);\n  }\n\n  override toString(): string {\n    return \"Tanh()\";\n  }\n}\n\n/**\n * Applies the Leaky Rectified Linear Unit (Leaky ReLU) activation.\n *\n * LeakyReLU(x) = max(alpha * x, x)\n *\n * @category Neural Network Layers\n */\nexport class LeakyReLU extends Module {\n  private readonly alpha: number;\n\n  constructor(alpha = 0.01) {\n    super();\n    this.alpha = alpha;\n  }\n\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) return input.leakyRelu(this.alpha);\n    return leakyReluOp(input, this.alpha);\n  }\n\n  override toString(): string {\n    return `LeakyReLU(alpha=${this.alpha})`;\n  }\n}\n\n/**\n * Applies the Exponential Linear Unit (ELU) activation.\n *\n * ELU(x) = x if x > 0, else alpha * (exp(x) - 1)\n *\n * @category Neural Network Layers\n */\nexport class ELU extends Module {\n  private readonly alpha: number;\n\n  constructor(alpha = 1.0) {\n    super();\n    // Store alpha parameter for negative values\n    // ELU can produce negative outputs, pushing mean activations closer to zero\n    this.alpha = alpha;\n  }\n\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) return input.elu(this.alpha);\n    return eluOp(input, this.alpha);\n  }\n\n  override toString(): string {\n    return `ELU(alpha=${this.alpha})`;\n  }\n}\n\n/**\n * Applies the Gaussian Error Linear Unit (GELU) activation.\n *\n * GELU(x) = x * Phi(x) where Phi is the CDF of standard normal distribution\n *\n * @category Neural Network Layers\n */\nexport class GELU extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) return input.gelu();\n    return geluOp(input);\n  }\n\n  override toString(): string {\n    return \"GELU()\";\n  }\n}\n\n/**\n * Applies the Softmax activation function.\n *\n * Softmax(x_i) = exp(x_i) / sum(exp(x_j))\n *\n * @category Neural Network Layers\n */\nexport class Softmax extends Module {\n  private readonly axis: Axis;\n\n  constructor(axis: Axis = -1) {\n    super();\n    // Store axis along which to compute softmax\n    // Default -1 means last axis (typical for classification)\n    this.axis = axis;\n  }\n\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) {\n      return gradSoftmax(input, normalizeAxis(this.axis, input.tensor.ndim));\n    }\n    return softmaxOp(input, this.axis);\n  }\n\n  override toString(): string {\n    return `Softmax(axis=${this.axis})`;\n  }\n}\n\n/**\n * Applies the Log Softmax activation function.\n *\n * LogSoftmax(x_i) = log(exp(x_i) / sum(exp(x_j)))\n *\n * @category Neural Network Layers\n */\nexport class LogSoftmax extends Module {\n  private readonly axis: Axis;\n\n  constructor(axis: Axis = -1) {\n    super();\n    // Store axis for log-softmax computation\n    // More numerically stable than log(softmax(x))\n    this.axis = axis;\n  }\n\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) {\n      return gradLogSoftmax(input, normalizeAxis(this.axis, input.tensor.ndim));\n    }\n    return logSoftmaxOp(input, this.axis);\n  }\n\n  override toString(): string {\n    return `LogSoftmax(axis=${this.axis})`;\n  }\n}\n\n/**\n * Applies the Softplus activation function.\n *\n * Softplus(x) = log(1 + exp(x))\n *\n * @category Neural Network Layers\n */\nexport class Softplus extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) {\n      return GradTensor.fromTensor(softplusOp(input.tensor), {\n        requiresGrad: false,\n      });\n    }\n    return softplusOp(input);\n  }\n\n  override toString(): string {\n    return \"Softplus()\";\n  }\n}\n\n/**\n * Applies the Swish activation function (also known as SiLU).\n *\n * Swish(x) = x * sigmoid(x)\n *\n * @category Neural Network Layers\n */\nexport class Swish extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) {\n      return GradTensor.fromTensor(swishOp(input.tensor), {\n        requiresGrad: false,\n      });\n    }\n    return swishOp(input);\n  }\n\n  override toString(): string {\n    return \"Swish()\";\n  }\n}\n\n/**\n * Applies the Mish activation function.\n *\n * Mish(x) = x * tanh(softplus(x))\n *\n * @category Neural Network Layers\n */\nexport class Mish extends Module {\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    if (input instanceof GradTensor) {\n      return GradTensor.fromTensor(mishOp(input.tensor), {\n        requiresGrad: false,\n      });\n    }\n    return mishOp(input);\n  }\n\n  override toString(): string {\n    return \"Mish()\";\n  }\n}\n","import { DTypeError, InvalidParameterError } from \"../../core\";\nimport { type AnyTensor, dropoutGrad, GradTensor } from \"../../ndarray\";\nimport { Module } from \"../module/Module\";\n\n/**\n * Applies Dropout regularization during training.\n *\n * **Mathematical Formulation:**\n * During training:\n * ```\n * y = x * mask / (1 - p)\n * ```\n * where mask is a binary tensor with probability (1-p) of being 1.\n *\n * During evaluation:\n * ```\n * y = x\n * ```\n *\n * **Purpose:**\n * - Prevents overfitting by randomly zeroing elements during training\n * - Forces network to learn redundant representations\n * - Improves generalization performance\n *\n * **Scaling:**\n * The output is scaled by 1/(1-p) during training to maintain expected value.\n * This is called \"inverted dropout\" and eliminates the need for scaling during inference.\n *\n * @example\n * ```ts\n * import { Dropout } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const dropout = new Dropout(0.5); // Drop 50% of neurons\n * const input = tensor([[1, 2, 3, 4]]);\n *\n * // Training mode: randomly zeros ~50% of elements\n * dropout.train();\n * const output = dropout.forward(input);\n *\n * // Evaluation mode: passes input unchanged\n * dropout.eval();\n * const output2 = dropout.forward(input); // Same as input\n * ```\n *\n * References:\n * - Dropout paper: https://jmlr.org/papers/v15/srivastava14a.html\n * - PyTorch Dropout: https://pytorch.org/docs/stable/generated/torch.nn.Dropout.html\n *\n * @category Neural Network Layers\n */\nexport class Dropout extends Module {\n  /** Probability of an element being zeroed (dropout rate) */\n  private readonly p: number;\n\n  /**\n   * Create a new Dropout layer.\n   *\n   * @param p - Probability of an element being zeroed (0 <= p < 1)\n   * @throws {InvalidParameterError} If p is not in valid range [0, 1)\n   */\n  constructor(p = 0.5) {\n    super();\n\n    // Validate dropout probability is in valid range\n    if (!Number.isFinite(p) || p < 0 || p >= 1) {\n      throw new InvalidParameterError(`Dropout probability must be in [0, 1), got ${p}`, \"p\", p);\n    }\n\n    this.p = p;\n  }\n\n  /**\n   * Forward pass: apply dropout during training, identity during evaluation.\n   *\n   * @param input - Input tensor of any shape (Tensor or GradTensor)\n   * @returns Output tensor with same shape as input\n   */\n  forward(input: AnyTensor): GradTensor {\n    // Convert to GradTensor if needed\n    const inputTensor = input instanceof GradTensor ? input : GradTensor.fromTensor(input);\n\n    if (inputTensor.dtype === \"string\") {\n      throw new DTypeError(\"Dropout does not support string dtype\");\n    }\n\n    // Use vectorized dropout implementation from autograd\n    // This handles training/eval mode and mask generation\n    return dropoutGrad(inputTensor, this.p, this.training);\n  }\n\n  /**\n   * Get string representation of the layer.\n   *\n   * @returns String representation with dropout probability\n   */\n  override toString(): string {\n    return `Dropout(p=${this.p})`;\n  }\n\n  /**\n   * Get the dropout probability.\n   */\n  get dropoutRate(): number {\n    return this.p;\n  }\n}\n","import { DTypeError, InvalidParameterError, ShapeError } from \"../../core\";\nimport type { Tensor } from \"../../ndarray\";\nimport {\n  add,\n  dot,\n  GradTensor,\n  mulScalar,\n  parameter,\n  randn,\n  reshape,\n  transpose,\n  zeros,\n} from \"../../ndarray\";\nimport { Module } from \"../module/Module\";\n\n/**\n * Applies a linear transformation to the incoming data: y = xA^T + b\n *\n * This is also known as a fully connected layer or dense layer.\n *\n * **Mathematical Formulation:**\n * ```\n * y = x * W^T + b\n * ```\n *\n * Where:\n * - x is the input tensor of shape (*, in_features)\n * - W is the weight matrix of shape (out_features, in_features)\n * - b is the bias vector of shape (out_features,)\n * - y is the output tensor of shape (*, out_features)\n *\n * **Shape Conventions:**\n * - Input: `(*, in_features)` where `*` means any number of leading dimensions\n *   - 1D: `(in_features)` → Output: `(out_features)`\n *   - 2D: `(batch, in_features)` → Output: `(batch, out_features)`\n *   - 3D: `(batch, seq_len, in_features)` → Output: `(batch, seq_len, out_features)`\n * - The last dimension must equal `in_features`\n * - All leading dimensions are preserved in the output\n *\n * **Parameters:**\n * - `inFeatures`: Size of each input sample\n * - `outFeatures`: Size of each output sample\n * - `bias`: If true, adds a learnable bias to the output\n *\n * **Attributes:**\n * - `weight`: Learnable weights of shape (out_features, in_features)\n * - `bias`: Learnable bias of shape (out_features,) if bias=true\n *\n * **Initialization:**\n * Uses Kaiming/He initialization: weights ~ N(0, sqrt(2/in_features))\n * Biases are initialized to zeros\n *\n * @example\n * ```ts\n * import { Linear } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * // Create a linear layer with 20 input features and 30 output features\n * const layer = new Linear(20, 30);\n *\n * // Forward pass\n * const input = tensor([[1, 2, ..., 20]]); // shape: (1, 20)\n * const output = layer.forward(input);     // shape: (1, 30)\n *\n * // Without bias\n * const layerNoBias = new Linear(10, 5, { bias: false });\n * ```\n *\n * References:\n * - PyTorch Linear: https://pytorch.org/docs/stable/generated/torch.nn.Linear.html\n * - Xavier/Glorot initialization: http://proceedings.mlr.press/v9/glorot10a.html\n *\n * @category Neural Network Layers\n */\nexport class Linear extends Module {\n  /** Weight matrix of shape (out_features, in_features) */\n  private weight: Tensor;\n  private weightParam: GradTensor;\n\n  /** Bias vector of shape (out_features,) */\n  private bias?: Tensor;\n  private biasParam?: GradTensor;\n\n  /** Number of input features */\n  private readonly inFeatures: number;\n\n  /** Number of output features */\n  private readonly outFeatures: number;\n\n  /** Whether this layer has a bias */\n  private readonly useBias: boolean;\n\n  /**\n   * Create a new Linear layer.\n   *\n   * @param inFeatures - Size of each input sample\n   * @param outFeatures - Size of each output sample\n   * @param options - Configuration options\n   * @param options.bias - If true, add learnable bias (default: true)\n   * @param options.dtype - Data type for weights (default: 'float32')\n   * @param options.device - Device to place tensors on (default: 'cpu')\n   */\n  constructor(\n    inFeatures: number,\n    outFeatures: number,\n    options: {\n      readonly bias?: boolean;\n      readonly dtype?: \"float32\" | \"float64\";\n      readonly device?: \"cpu\" | \"webgpu\" | \"wasm\";\n    } = {}\n  ) {\n    // Call parent Module constructor to initialize base class\n    super();\n\n    // Validate dimensions\n    if (inFeatures <= 0 || !Number.isInteger(inFeatures)) {\n      throw new InvalidParameterError(\n        \"inFeatures must be a positive integer\",\n        \"inFeatures\",\n        inFeatures\n      );\n    }\n    if (outFeatures <= 0 || !Number.isInteger(outFeatures)) {\n      throw new InvalidParameterError(\n        \"outFeatures must be a positive integer\",\n        \"outFeatures\",\n        outFeatures\n      );\n    }\n\n    // Store layer dimensions for validation and access\n    this.inFeatures = inFeatures;\n    this.outFeatures = outFeatures;\n    // Default to using bias unless explicitly disabled\n    this.useBias = options.bias ?? true;\n\n    // Initialize weights using Kaiming initialization (He initialization)\n    // This initialization is optimal for ReLU activations\n    // Standard deviation: sqrt(2 / fan_in) where fan_in = inFeatures\n    const stdDev = Math.sqrt(2.0 / inFeatures);\n\n    // Create weight matrix with shape (out_features, in_features)\n    // Transposed storage allows efficient matrix multiplication: y = x * W^T\n    const weightTensor = randn([outFeatures, inFeatures], {\n      dtype: options.dtype ?? \"float32\",\n      device: options.device ?? \"cpu\",\n    });\n\n    // Scale the randomly initialized weights by the computed standard deviation\n    // This ensures proper gradient flow during backpropagation\n    const scaledWeight = mulScalar(weightTensor, stdDev);\n    this.weightParam = parameter(scaledWeight);\n    this.weight = this.weightParam.tensor;\n\n    // Register weight as a trainable parameter for optimizer access\n    this.registerParameter(\"weight\", this.weightParam);\n\n    // Initialize bias to zeros if enabled (common practice)\n    // Bias allows the layer to shift the output independently of input\n    if (this.useBias) {\n      // Bias has shape (out_features,) - one value per output neuron\n      const biasTensor = zeros([outFeatures], {\n        dtype: options.dtype ?? \"float32\",\n        device: options.device ?? \"cpu\",\n      });\n      this.biasParam = parameter(biasTensor);\n      this.bias = this.biasParam.tensor;\n      // Register bias as a trainable parameter\n      this.registerParameter(\"bias\", this.biasParam);\n    }\n  }\n\n  /**\n   * Forward pass: compute y = x * W^T + b\n   *\n   * @param input - Input tensor of shape (*, in_features)\n   * @returns Output tensor of shape (*, out_features)\n   * @throws {ShapeError} If input shape is invalid\n   * @throws {DTypeError} If input dtype is unsupported\n   */\n  forward(input: GradTensor): GradTensor;\n  forward(input: Tensor): Tensor;\n  forward(input: Tensor | GradTensor): Tensor | GradTensor {\n    const inputTensor = input instanceof GradTensor ? input.tensor : input;\n\n    if (inputTensor.dtype === \"string\") {\n      throw new DTypeError(\"Linear layer does not support string dtype\");\n    }\n\n    // Validate input dimensionality - must be at least 1D\n    // 0D (scalar) inputs are not valid for linear transformations\n    if (inputTensor.ndim < 1) {\n      throw new ShapeError(`Linear layer expects at least 1D input; got ndim=${inputTensor.ndim}`);\n    }\n\n    // Extract the last dimension size (number of features)\n    // For input shape (batch, seq_len, features), this gets 'features'\n    const inputFeatures = inputTensor.shape[inputTensor.shape.length - 1] ?? 0;\n\n    // Validate that input features match the layer's expected input size\n    if (inputFeatures !== this.inFeatures) {\n      throw new ShapeError(\n        `Linear layer expects ${this.inFeatures} input features; got ${inputFeatures}`\n      );\n    }\n\n    // Compute the linear transformation: y = x * W^T + b\n    // Weight is stored as (out_features, in_features), so we transpose it\n    // This allows efficient computation: (batch, in_features) @ (in_features, out_features)\n\n    // Check if input is a 1D vector (no batch dimension)\n    const isVectorInput = inputTensor.ndim === 1;\n\n    // Calculate total batch size (handles multi-dimensional batches)\n    // For shape (batch, seq, features): batchSize = batch * seq\n    const batchSize = inputTensor.size / this.inFeatures;\n\n    // Reshape input to 2D: (batchSize, inFeatures) for matrix multiplication\n    const outputShape = isVectorInput\n      ? [this.outFeatures]\n      : [...inputTensor.shape.slice(0, -1), this.outFeatures];\n\n    if (input instanceof GradTensor) {\n      const input2d = input.reshape([batchSize, this.inFeatures]);\n      const output2d = input2d.matmul(this.weightParam.transpose());\n      let output = output2d.reshape(outputShape);\n      if (this.useBias && this.biasParam) {\n        output = output.add(this.biasParam);\n      }\n      return output;\n    }\n\n    const input2d = reshape(inputTensor, [batchSize, this.inFeatures]);\n\n    // Perform matrix multiplication: (batchSize, inFeatures) @ (inFeatures, outFeatures)\n    // Result shape: (batchSize, outFeatures)\n    const output2d = dot(input2d, transpose(this.weight));\n\n    // Restore original batch shape, replacing last dimension with outFeatures\n    // If input was 1D, output should be 1D; if (batch, features), output is (batch, outFeatures)\n    const output = reshape(output2d, outputShape);\n\n    // Add bias term if enabled\n    if (this.useBias && this.bias) {\n      // Broadcasting automatically handles all batch dimensions\n      // Bias shape (outFeatures,) broadcasts to (..., outFeatures)\n      return add(output, this.bias);\n    }\n\n    // Return output without bias\n    return output;\n  }\n\n  /**\n   * Get extra representation string for this layer.\n   *\n   * @returns String representation of layer parameters\n   */\n  override toString(): string {\n    const biasStr = this.useBias ? \"bias=true\" : \"bias=false\";\n    return `Linear(in_features=${this.inFeatures}, out_features=${this.outFeatures}, ${biasStr})`;\n  }\n\n  /**\n   * Get the weight matrix.\n   *\n   * @returns Weight tensor of shape (out_features, in_features)\n   */\n  getWeight(): Tensor {\n    return this.weight;\n  }\n\n  /**\n   * Get the bias vector.\n   *\n   * @returns Bias tensor of shape (out_features,) or undefined if no bias\n   */\n  getBias(): Tensor | undefined {\n    return this.bias;\n  }\n\n  /**\n   * Get the number of input features.\n   */\n  get inputSize(): number {\n    return this.inFeatures;\n  }\n\n  /**\n   * Get the number of output features.\n   */\n  get outputSize(): number {\n    return this.outFeatures;\n  }\n}\n","import {\n  DTypeError,\n  dtypeToTypedArrayCtor,\n  ensureNumericDType,\n  getBigIntElement,\n  getNumericElement,\n  InvalidParameterError,\n  ShapeError,\n} from \"../../core\";\nimport {\n  type AnyTensor,\n  GradTensor,\n  noGrad,\n  ones,\n  parameter,\n  varianceGrad,\n  zeros,\n} from \"../../ndarray\";\nimport { isContiguous, offsetFromFlatIndex } from \"../../ndarray/tensor/strides\";\nimport { computeStrides, Tensor as TensorClass } from \"../../ndarray/tensor/Tensor\";\nimport { Module } from \"../module/Module\";\n\nfunction toContiguousTensor(t: TensorClass): TensorClass {\n  if (isContiguous(t.shape, t.strides)) {\n    return t;\n  }\n  if (t.dtype === \"string\") {\n    throw new DTypeError(\"Normalization does not support string dtype\");\n  }\n  const Ctor = dtypeToTypedArrayCtor(t.dtype);\n  const out = new Ctor(t.size);\n  const logicalStrides = computeStrides(t.shape);\n  const data = t.data;\n\n  if (Array.isArray(data)) {\n    throw new DTypeError(\"Normalization does not support string dtype\");\n  }\n\n  if (data instanceof BigInt64Array) {\n    if (!(out instanceof BigInt64Array)) {\n      throw new DTypeError(\"Expected int64 output buffer for int64 tensor\");\n    }\n    for (let i = 0; i < t.size; i++) {\n      const offset = offsetFromFlatIndex(i, logicalStrides, t.strides, t.offset);\n      out[i] = getBigIntElement(data, offset);\n    }\n  } else {\n    if (out instanceof BigInt64Array) {\n      throw new DTypeError(\"Unexpected int64 output buffer for numeric tensor\");\n    }\n    for (let i = 0; i < t.size; i++) {\n      const offset = offsetFromFlatIndex(i, logicalStrides, t.strides, t.offset);\n      out[i] = getNumericElement(data, offset);\n    }\n  }\n\n  return TensorClass.fromTypedArray({\n    data: out,\n    shape: t.shape,\n    dtype: t.dtype,\n    device: t.device,\n  });\n}\n\n/**\n * Batch Normalization layer.\n *\n * Normalizes the input over the batch dimension for faster and more stable training.\n *\n * **Formula**: y = (x - E[x]) / sqrt(Var[x] + eps) * gamma + beta\n *\n * During training, uses batch statistics. During evaluation, uses running statistics\n * unless `trackRunningStats=false`, in which case batch statistics are always used.\n *\n * @example\n * ```ts\n * import { BatchNorm1d } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const bn = new BatchNorm1d(10);\n * const x = tensor([[1, 2, 3], [4, 5, 6]]);\n * const y = bn.forward(x);\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm1d.html | PyTorch BatchNorm1d}\n */\nexport class BatchNorm1d extends Module {\n  private readonly numFeatures: number;\n  private readonly eps: number;\n  private readonly momentum: number;\n  private readonly affine: boolean;\n  private readonly trackRunningStats: boolean;\n\n  private gamma?: GradTensor;\n  private beta?: GradTensor;\n  private runningMean: GradTensor;\n  private runningVar: GradTensor;\n\n  constructor(\n    numFeatures: number,\n    options: {\n      readonly eps?: number;\n      readonly momentum?: number;\n      readonly affine?: boolean;\n      readonly trackRunningStats?: boolean;\n    } = {}\n  ) {\n    super();\n    if (\n      !Number.isFinite(numFeatures) ||\n      numFeatures <= 0 ||\n      Math.trunc(numFeatures) !== numFeatures\n    ) {\n      throw new InvalidParameterError(\n        \"numFeatures must be a positive integer\",\n        \"numFeatures\",\n        numFeatures\n      );\n    }\n    this.numFeatures = numFeatures;\n    this.eps = options.eps ?? 1e-5;\n    if (!Number.isFinite(this.eps) || this.eps <= 0) {\n      throw new InvalidParameterError(\"eps must be a positive number\", \"eps\", this.eps);\n    }\n    this.momentum = options.momentum ?? 0.1;\n    if (!Number.isFinite(this.momentum) || this.momentum < 0 || this.momentum > 1) {\n      throw new InvalidParameterError(\n        \"momentum must be in range [0, 1]\",\n        \"momentum\",\n        this.momentum\n      );\n    }\n    this.affine = options.affine ?? true;\n    this.trackRunningStats = options.trackRunningStats ?? true;\n\n    if (this.affine) {\n      const gamma = ones([numFeatures]);\n      const beta = zeros([numFeatures]);\n      this.gamma = parameter(gamma);\n      this.beta = parameter(beta);\n      this.registerParameter(\"weight\", this.gamma);\n      this.registerParameter(\"bias\", this.beta);\n    }\n\n    // buffers\n    this.runningMean = GradTensor.fromTensor(zeros([numFeatures]), {\n      requiresGrad: false,\n    });\n    this.runningVar = GradTensor.fromTensor(ones([numFeatures]), {\n      requiresGrad: false,\n    });\n\n    if (this.trackRunningStats) {\n      this.registerBuffer(\"running_mean\", this.runningMean.tensor);\n      this.registerBuffer(\"running_var\", this.runningVar.tensor);\n    }\n  }\n\n  forward(x: AnyTensor): GradTensor {\n    const input = x instanceof GradTensor ? x : GradTensor.fromTensor(x);\n\n    const inputDtype = input.dtype;\n    if (inputDtype === \"string\") {\n      throw new DTypeError(\"BatchNorm1d does not support string dtype\");\n    }\n\n    if (input.ndim !== 2 && input.ndim !== 3) {\n      throw new ShapeError(`BatchNorm1d expects 2D or 3D input; got ndim=${input.ndim}`);\n    }\n\n    const nFeatures = input.shape[1] ?? 0;\n    if (nFeatures !== this.numFeatures) {\n      throw new ShapeError(`Expected ${this.numFeatures} features, got ${nFeatures}`);\n    }\n\n    const useBatchStats = this.training || !this.trackRunningStats;\n\n    let mean: GradTensor;\n    let varVal: GradTensor;\n\n    // We need to reduce over all dims except channel (dim 1).\n    // For 2D (N, C): reduce over 0.\n    // For 3D (N, C, L): reduce over 0 and 2.\n    // To use single variance op, we reshape to (N*L, C) or (C, N*L).\n    // Better to reshape to (BatchTotal, C) and reduce over 0.\n    // For 2D: already (N, C).\n    // For 3D: permute to (N, L, C) -> reshape (N*L, C).\n\n    let inputReshaped = input;\n    if (input.ndim === 3) {\n      // (N, C, L) -> (N, L, C) -> (N*L, C)\n      const batch = input.shape[0] ?? 0;\n      const length = input.shape[2] ?? 0;\n      const flat = batch * length;\n      const numericInputDtype = ensureNumericDType(inputDtype, \"BatchNorm1d\");\n      inputReshaped = input\n        .transpose([0, 2, 1])\n        .mul(GradTensor.scalar(1, { dtype: numericInputDtype }))\n        .reshape([flat, nFeatures]);\n    }\n\n    if (useBatchStats) {\n      if (inputReshaped.shape[0] === 0) {\n        throw new InvalidParameterError(\n          \"BatchNorm requires at least one element\",\n          \"input\",\n          input.shape\n        );\n      }\n\n      // Compute batch statistics (biased variance for normalization, per PyTorch convention)\n      mean = inputReshaped.mean(0);\n      varVal = varianceGrad(inputReshaped, 0, 0);\n\n      if (this.trackRunningStats) {\n        noGrad(() => {\n          // Use unbiased variance (Bessel's correction) for running stats update\n          const n = inputReshaped.shape[0] ?? 0;\n          const unbiasedVar =\n            n > 1 ? varianceGrad(inputReshaped, 0, 1) : varianceGrad(inputReshaped, 0, 0);\n\n          const m = this.momentum;\n          const statsDtype = this.runningMean.dtype;\n          if (statsDtype === \"string\") {\n            throw new DTypeError(\"BatchNorm running statistics must be numeric\");\n          }\n          const oneMinusM = GradTensor.scalar(1 - m, { dtype: statsDtype });\n          const mScalar = GradTensor.scalar(m, { dtype: statsDtype });\n\n          // Exponential moving average: running = (1 - momentum) * running + momentum * batch\n          const newMean = this.runningMean.mul(oneMinusM).add(mean.mul(mScalar));\n          const newVar = this.runningVar.mul(oneMinusM).add(unbiasedVar.mul(mScalar));\n\n          // Detach to prevent graph history in running stats\n          this.runningMean = GradTensor.fromTensor(newMean.tensor, {\n            requiresGrad: false,\n          });\n          this.runningVar = GradTensor.fromTensor(newVar.tensor, {\n            requiresGrad: false,\n          });\n\n          // Re-register buffers to keep Module._buffers map in sync\n          this.registerBuffer(\"running_mean\", this.runningMean.tensor);\n          this.registerBuffer(\"running_var\", this.runningVar.tensor);\n        });\n      }\n    } else {\n      mean = this.runningMean;\n      varVal = this.runningVar;\n    }\n\n    // Normalize: (x - mean) / sqrt(var + eps)\n    // We need to broadcast mean/var to input shape.\n    // Input: (N, C) or (N, C, L). Mean/Var: (C).\n    // For (N, C), broadcasting (C) works (last dim matches).\n    // For (N, C, L), we need (1, C, 1).\n\n    let meanBroadcast = mean;\n    let varBroadcast = varVal;\n\n    if (input.ndim === 3) {\n      meanBroadcast = mean.reshape([1, nFeatures, 1]);\n      varBroadcast = varVal.reshape([1, nFeatures, 1]);\n    } else {\n      meanBroadcast = mean.reshape([1, nFeatures]);\n      varBroadcast = varVal.reshape([1, nFeatures]);\n    }\n\n    const epsTensor = GradTensor.scalar(this.eps, { dtype: inputDtype });\n    const denom = varBroadcast.add(epsTensor).sqrt();\n    let out = input.sub(meanBroadcast).div(denom);\n\n    if (this.affine && this.gamma && this.beta) {\n      let gammaB = this.gamma;\n      let betaB = this.beta;\n      if (input.ndim === 3) {\n        gammaB = this.gamma.reshape([1, nFeatures, 1]);\n        betaB = this.beta.reshape([1, nFeatures, 1]);\n      } else {\n        gammaB = this.gamma.reshape([1, nFeatures]);\n        betaB = this.beta.reshape([1, nFeatures]);\n      }\n      out = out.mul(gammaB).add(betaB);\n    }\n\n    return out;\n  }\n\n  override toString(): string {\n    return `BatchNorm1d(${this.numFeatures}, eps=${this.eps}, momentum=${this.momentum}, affine=${this.affine})`;\n  }\n}\n\n/**\n * Layer Normalization.\n *\n * Normalizes across the feature dimensions (trailing dimensions specified by `normalizedShape`)\n * for each sample independently. Unlike BatchNorm, LayerNorm works the same way during training\n * and evaluation.\n *\n * **Formula**: y = (x - E[x]) / sqrt(Var[x] + eps) * gamma + beta\n *\n * @example\n * ```ts\n * import { LayerNorm } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const ln = new LayerNorm([10]);\n * const x = tensor([[1, 2, 3]]);\n * const y = ln.forward(x);\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html | PyTorch LayerNorm}\n */\nexport class LayerNorm extends Module {\n  private readonly normalizedShape: readonly number[];\n  private readonly eps: number;\n  private readonly elementwiseAffine: boolean;\n\n  private gamma?: GradTensor;\n  private beta?: GradTensor;\n\n  constructor(\n    normalizedShape: number | readonly number[],\n    options: {\n      readonly eps?: number;\n      readonly elementwiseAffine?: boolean;\n    } = {}\n  ) {\n    super();\n    this.normalizedShape =\n      typeof normalizedShape === \"number\" ? [normalizedShape] : Array.from(normalizedShape);\n\n    if (this.normalizedShape.length === 0) {\n      throw new InvalidParameterError(\n        \"normalizedShape must contain at least one dimension\",\n        \"normalizedShape\",\n        normalizedShape\n      );\n    }\n\n    for (const dim of this.normalizedShape) {\n      if (!Number.isFinite(dim) || dim <= 0 || Math.trunc(dim) !== dim) {\n        throw new InvalidParameterError(\n          \"All dimensions in normalizedShape must be positive integers\",\n          \"normalizedShape\",\n          normalizedShape\n        );\n      }\n    }\n\n    this.eps = options.eps ?? 1e-5;\n    if (!Number.isFinite(this.eps) || this.eps <= 0) {\n      throw new InvalidParameterError(\"eps must be a positive number\", \"eps\", this.eps);\n    }\n\n    this.elementwiseAffine = options.elementwiseAffine ?? true;\n\n    if (this.elementwiseAffine) {\n      this.gamma = parameter(ones(this.normalizedShape));\n      this.beta = parameter(zeros(this.normalizedShape));\n      this.registerParameter(\"weight\", this.gamma);\n      this.registerParameter(\"bias\", this.beta);\n    }\n  }\n\n  forward(x: AnyTensor): GradTensor {\n    const input = x instanceof GradTensor ? x : GradTensor.fromTensor(x);\n\n    const inputDtype = input.dtype;\n    if (inputDtype === \"string\") {\n      throw new DTypeError(\"LayerNorm does not support string dtype\");\n    }\n\n    let workingInput = input;\n    if (!isContiguous(input.tensor.shape, input.tensor.strides)) {\n      const contiguous = toContiguousTensor(input.tensor);\n      workingInput = GradTensor.fromTensor(contiguous, {\n        requiresGrad: input.requiresGrad,\n      });\n    }\n\n    // Check if input shape ends with normalizedShape\n    const inputShape = workingInput.shape;\n    const normShape = this.normalizedShape;\n    if (normShape.length > inputShape.length) {\n      throw new ShapeError(`Input shape ${inputShape} too small for normalizedShape ${normShape}`);\n    }\n\n    // Check suffix\n    const suffixStart = inputShape.length - normShape.length;\n    for (let i = 0; i < normShape.length; i++) {\n      if (inputShape[suffixStart + i] !== normShape[i]) {\n        throw new ShapeError(\n          `Input shape ${inputShape} does not end with normalizedShape ${normShape}`\n        );\n      }\n    }\n\n    // We need to flatten the normalized dimensions to calculate mean/var over them.\n    // Dimensions to reduce: [suffixStart, ..., inputShape.length - 1]\n    // We can reshape input to (..., Product(normShape)).\n    // Then reduce over last dimension.\n\n    const outerDims = inputShape.slice(0, suffixStart);\n    const normSize = normShape.reduce((a, b) => a * b, 1);\n\n    const flattenedShape = [...outerDims, normSize];\n    const inputReshaped = workingInput.reshape(flattenedShape);\n\n    // Mean and Var over last dim (-1)\n    const mean = inputReshaped.mean(-1, true); // Keep dims to facilitate broadcasting (..., 1)\n    const varVal = varianceGrad(inputReshaped, -1, 0); // Biased variance for normalization\n    // varianceGrad returns tensor with reduced dim removed?\n    // Wait, varianceGrad(..., axis) removes the axis unless keepdims?\n    // My implementation of `varianceGrad` calls `mean` and `sum`.\n    // `mean(axis)` removes dim unless `keepdims=true`.\n    // `varianceGrad` implementation in `autograd`:\n    // It doesn't take `keepdims` param. It reduces.\n    // So `varVal` has shape (...,).\n    // We need to reshape it to (..., 1) for broadcasting.\n\n    const varReshaped = varVal.reshape(mean.shape); // mean has kept dims\n\n    // Normalize\n    const epsTensor = GradTensor.scalar(this.eps, { dtype: inputDtype });\n    const denom = varReshaped.add(epsTensor).sqrt();\n    const normalizedReshaped = inputReshaped.sub(mean).div(denom);\n\n    // Reshape back to original shape\n    let out = normalizedReshaped.reshape(inputShape);\n\n    // Apply affine\n    if (this.elementwiseAffine && this.gamma && this.beta) {\n      // gamma/beta shape: normShape\n      // input shape: (..., normShape)\n      // Broadcasting works automatically since trailing dims match.\n      out = out.mul(this.gamma).add(this.beta);\n    }\n\n    return out;\n  }\n\n  override toString(): string {\n    return `LayerNorm(${this.normalizedShape}, eps=${this.eps}, elementwise_affine=${this.elementwiseAffine})`;\n  }\n}\n","import { DTypeError, ensureNumericDType, InvalidParameterError, ShapeError } from \"../../core\";\nimport {\n  type AnyTensor,\n  dropoutGrad,\n  GradTensor,\n  mulScalar,\n  parameter,\n  randn,\n  softmaxGrad,\n  zeros,\n} from \"../../ndarray\";\nimport { Module } from \"../module/Module\";\nimport { Dropout } from \"./dropout\";\nimport { Linear } from \"./linear\";\nimport { LayerNorm } from \"./normalization\";\n\n/**\n * Multi-Head Attention mechanism.\n *\n * Allows the model to jointly attend to information from different representation\n * subspaces at different positions. This is the core building block of Transformers.\n *\n * **Mathematical Formulation**:\n * ```\n * Attention(Q, K, V) = softmax(Q * K^T / sqrt(d_k)) * V\n * MultiHead(Q, K, V) = Concat(head_1, ..., head_h) * W_O\n * where head_i = Attention(Q * W_Q^i, K * W_K^i, V * W_V^i)\n * ```\n *\n * @example\n * ```ts\n * import { MultiheadAttention } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const mha = new MultiheadAttention(512, 8);\n * const x = tensor([[/* ... sequence data ... *\\/]]);\n * const output = mha.forward(x, x, x);\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.MultiheadAttention.html | PyTorch MultiheadAttention}\n * @see Vaswani et al. (2017) \"Attention Is All You Need\"\n */\nexport class MultiheadAttention extends Module {\n  /** Embedding dimension */\n  private readonly embedDim: number;\n\n  /** Number of attention heads */\n  private readonly numHeads: number;\n\n  /** Dimension of each head */\n  private readonly headDim: number;\n\n  /** Scaling factor for dot product attention */\n  private readonly scale: number;\n\n  /** Whether to add bias to projections */\n  private readonly useBias: boolean;\n\n  /** Dropout probability applied to attention weights */\n  private readonly dropout: number;\n\n  /** Query projection weights (embedDim, embedDim) */\n  private wQ: GradTensor;\n  private bQ?: GradTensor;\n\n  /** Key projection weights (embedDim, embedDim) */\n  private wK: GradTensor;\n  private bK?: GradTensor;\n\n  /** Value projection weights (embedDim, embedDim) */\n  private wV: GradTensor;\n  private bV?: GradTensor;\n\n  /** Output projection weights (embedDim, embedDim) */\n  private wO: GradTensor;\n  private bO?: GradTensor;\n\n  /**\n   * Create a new MultiheadAttention layer.\n   *\n   * @param embedDim - Total dimension of the model (must be divisible by numHeads)\n   * @param numHeads - Number of parallel attention heads\n   * @param options - Configuration options\n   * @param options.bias - Whether to add bias to projections (default: true)\n   * @param options.dropout - Dropout probability applied to attention weights (default: 0.0)\n   */\n  constructor(\n    embedDim: number,\n    numHeads: number,\n    options: {\n      readonly bias?: boolean;\n      readonly dropout?: number;\n    } = {}\n  ) {\n    super();\n\n    if (!Number.isInteger(embedDim) || embedDim <= 0) {\n      throw new InvalidParameterError(\"embedDim must be a positive integer\", \"embedDim\", embedDim);\n    }\n    if (!Number.isInteger(numHeads) || numHeads <= 0) {\n      throw new InvalidParameterError(\"numHeads must be a positive integer\", \"numHeads\", numHeads);\n    }\n    if (embedDim % numHeads !== 0) {\n      throw new InvalidParameterError(\n        `embedDim (${embedDim}) must be divisible by numHeads (${numHeads})`,\n        \"embedDim\",\n        embedDim\n      );\n    }\n\n    const dropout = options.dropout ?? 0.0;\n    if (!Number.isFinite(dropout) || dropout < 0 || dropout >= 1) {\n      throw new InvalidParameterError(\"dropout must be in [0, 1)\", \"dropout\", dropout);\n    }\n\n    this.embedDim = embedDim;\n    this.numHeads = numHeads;\n    this.headDim = embedDim / numHeads;\n    this.scale = Math.sqrt(this.headDim);\n    this.useBias = options.bias ?? true;\n    this.dropout = dropout;\n\n    // Initialize projection weights using Xavier/Glorot initialization\n    const stdDev = Math.sqrt(2.0 / (embedDim + embedDim));\n\n    // Query, Key, Value projections\n    // We use GradTensor parameter directly\n    this.wQ = parameter(mulScalar(randn([embedDim, embedDim]), stdDev));\n    this.wK = parameter(mulScalar(randn([embedDim, embedDim]), stdDev));\n    this.wV = parameter(mulScalar(randn([embedDim, embedDim]), stdDev));\n    this.wO = parameter(mulScalar(randn([embedDim, embedDim]), stdDev));\n\n    this.registerParameter(\"in_proj_weight_q\", this.wQ);\n    this.registerParameter(\"in_proj_weight_k\", this.wK);\n    this.registerParameter(\"in_proj_weight_v\", this.wV);\n    this.registerParameter(\"out_proj_weight\", this.wO);\n\n    if (this.useBias) {\n      this.bQ = parameter(zeros([embedDim]));\n      this.bK = parameter(zeros([embedDim]));\n      this.bV = parameter(zeros([embedDim]));\n      this.bO = parameter(zeros([embedDim]));\n\n      this.registerParameter(\"in_proj_bias_q\", this.bQ);\n      this.registerParameter(\"in_proj_bias_k\", this.bK);\n      this.registerParameter(\"in_proj_bias_v\", this.bV);\n      this.registerParameter(\"out_proj_bias\", this.bO);\n    }\n  }\n\n  /**\n   * Forward pass of multi-head attention.\n   *\n   * @param query - Query tensor of shape (batch, seqLen, embedDim)\n   * @param key - Key tensor of shape (batch, seqLen, embedDim)\n   * @param value - Value tensor of shape (batch, seqLen, embedDim)\n   * @returns Output tensor of same shape as query\n   */\n  forward(...inputs: AnyTensor[]): GradTensor {\n    if (inputs.length < 1 || inputs.length > 3) {\n      throw new InvalidParameterError(\n        \"MultiheadAttention.forward expects 1 to 3 input tensors\",\n        \"inputs\",\n        inputs.length\n      );\n    }\n\n    const queryInput = inputs[0];\n    if (queryInput === undefined) {\n      throw new InvalidParameterError(\"Query tensor is required\", \"query\", queryInput);\n    }\n\n    // Auto-convert to GradTensor\n    const query = queryInput instanceof GradTensor ? queryInput : GradTensor.fromTensor(queryInput);\n\n    const keyInput = inputs[1] ?? queryInput;\n    const key = keyInput instanceof GradTensor ? keyInput : GradTensor.fromTensor(keyInput);\n\n    const valueInput = inputs[2] ?? queryInput;\n    const value = valueInput instanceof GradTensor ? valueInput : GradTensor.fromTensor(valueInput);\n\n    if (query.dtype === \"string\") throw new DTypeError(\"String tensors are not supported\");\n    if (query.ndim !== key.ndim || query.ndim !== value.ndim) {\n      throw new ShapeError(\"query, key, and value must have same rank\");\n    }\n    if (query.ndim !== 2 && query.ndim !== 3) {\n      throw new ShapeError(`Query must be 2D or 3D; got ndim=${query.ndim}`);\n    }\n    if (key.ndim !== 2 && key.ndim !== 3) {\n      throw new ShapeError(`Key must be 2D or 3D; got ndim=${key.ndim}`);\n    }\n    if (value.ndim !== 2 && value.ndim !== 3) {\n      throw new ShapeError(`Value must be 2D or 3D; got ndim=${value.ndim}`);\n    }\n\n    // Shape convention: (Batch, SeqLen, EmbedDim) for 3D inputs.\n    // If 2D (SeqLen, EmbedDim), we treat as (1, SeqLen, EmbedDim).\n\n    let q = query;\n    let k = key;\n    let v = value;\n\n    if (q.ndim === 2) q = q.reshape([1, q.shape[0] ?? 0, q.shape[1] ?? 0]);\n    if (k.ndim === 2) k = k.reshape([1, k.shape[0] ?? 0, k.shape[1] ?? 0]);\n    if (v.ndim === 2) v = v.reshape([1, v.shape[0] ?? 0, v.shape[1] ?? 0]);\n\n    const batchSize = q.shape[0] ?? 0;\n    const seqLenQ = q.shape[1] ?? 0;\n    const seqLenK = k.shape[1] ?? 0;\n    const seqLenV = v.shape[1] ?? 0;\n    const embedDim = q.shape[2] ?? 0;\n\n    if (embedDim !== this.embedDim) {\n      throw new ShapeError(`Query embedDim mismatch: expected ${this.embedDim}, got ${embedDim}`);\n    }\n    if (k.shape[2] !== this.embedDim) {\n      throw new ShapeError(`Key embedDim mismatch: expected ${this.embedDim}, got ${k.shape[2]}`);\n    }\n    if (v.shape[2] !== this.embedDim) {\n      throw new ShapeError(`Value embedDim mismatch: expected ${this.embedDim}, got ${v.shape[2]}`);\n    }\n    if (k.shape[0] !== batchSize || v.shape[0] !== batchSize) {\n      throw new ShapeError(\n        `batch size mismatch: query=${batchSize}, key=${k.shape[0]}, value=${v.shape[0]}`\n      );\n    }\n    if (seqLenK !== seqLenV) {\n      throw new ShapeError(`Key/value sequence length mismatch: key=${seqLenK}, value=${seqLenV}`);\n    }\n\n    // Linear projections\n    // Q * WQ^T + bQ\n    // (B, L, E) @ (E, E) -> (B, L, E)\n    let Q = q.matmul(this.wQ.transpose());\n    if (this.bQ) Q = Q.add(this.bQ);\n\n    let K = k.matmul(this.wK.transpose());\n    if (this.bK) K = K.add(this.bK);\n\n    let V = v.matmul(this.wV.transpose());\n    if (this.bV) V = V.add(this.bV);\n\n    // Split heads\n    // (B, L, E) -> (B, L, H, D) -> (B, H, L, D)\n    const H = this.numHeads;\n    const D = this.headDim;\n\n    Q = Q.reshape([batchSize, seqLenQ, H, D]).transpose([0, 2, 1, 3]);\n    K = K.reshape([batchSize, seqLenK, H, D]).transpose([0, 2, 1, 3]);\n    V = V.reshape([batchSize, seqLenV, H, D]).transpose([0, 2, 1, 3]);\n\n    // Scaled Dot-Product Attention\n    // Scores = Q @ K^T / sqrt(D)\n    // (B, H, Lq, D) @ (B, H, D, Lk) -> (B, H, Lq, Lk)\n    let scores = Q.matmul(K.transpose([0, 1, 3, 2]));\n    scores = scores.div(GradTensor.scalar(this.scale));\n\n    // Softmax\n    let attn = softmaxGrad(scores, -1);\n\n    // Dropout\n    attn = dropoutGrad(attn, this.dropout, this.training);\n\n    // Weighted Sum\n    // (B, H, Lq, Lk) @ (B, H, Lv, D) -> (B, H, Lq, D)\n    // Note: Lk == Lv usually\n    const context = attn.matmul(V);\n\n    // Concat heads\n    // (B, H, Lq, D) -> (B, Lq, H, D) -> (B, Lq, E)\n    const contextDtype = ensureNumericDType(context.dtype, \"MultiheadAttention\");\n    const contextReshaped = context\n      .transpose([0, 2, 1, 3])\n      .mul(GradTensor.scalar(1, { dtype: contextDtype }))\n      .reshape([batchSize, seqLenQ, this.embedDim]);\n\n    // Output projection\n    let output = contextReshaped.matmul(this.wO.transpose());\n    if (this.bO) output = output.add(this.bO);\n\n    // If input was 2D, squeeze back\n    if (query.ndim === 2) {\n      output = output.reshape([seqLenQ, this.embedDim]);\n    }\n\n    return output;\n  }\n\n  override toString(): string {\n    return `MultiheadAttention(embed_dim=${this.embedDim}, num_heads=${this.numHeads})`;\n  }\n}\n\n/**\n * Transformer Encoder Layer.\n *\n * A single layer of the Transformer encoder, consisting of:\n * 1. Multi-head self-attention\n * 2. Add & Norm (residual connection + layer normalization)\n * 3. Feed-forward network (FFN)\n * 4. Add & Norm\n *\n * @example\n * ```ts\n * import { TransformerEncoderLayer } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const layer = new TransformerEncoderLayer(512, 8, 2048);\n * const x = tensor([[/* sequence data *\\/]]);\n * const output = layer.forward(x);\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.TransformerEncoderLayer.html | PyTorch TransformerEncoderLayer}\n */\nexport class TransformerEncoderLayer extends Module {\n  private readonly dModel: number;\n  private readonly nHead: number;\n  private readonly dFF: number;\n\n  private readonly selfAttn: MultiheadAttention;\n  private readonly linear1: Linear;\n  private readonly linear2: Linear;\n  private readonly norm1: LayerNorm;\n  private readonly norm2: LayerNorm;\n\n  private readonly dropout: number;\n  // We use functional dropout in forward, or could use Dropout module.\n  // Using Dropout module is cleaner.\n  private readonly dropout1: Dropout;\n  private readonly dropout2: Dropout;\n  private readonly dropout3: Dropout;\n\n  constructor(\n    dModel: number,\n    nHead: number,\n    dFF: number,\n    options: {\n      readonly dropout?: number;\n      readonly eps?: number;\n    } = {}\n  ) {\n    super();\n\n    if (!Number.isInteger(dModel) || dModel <= 0) {\n      throw new InvalidParameterError(\"dModel must be a positive integer\", \"dModel\", dModel);\n    }\n    if (!Number.isInteger(nHead) || nHead <= 0) {\n      throw new InvalidParameterError(\"nHead must be a positive integer\", \"nHead\", nHead);\n    }\n    if (dModel % nHead !== 0) {\n      throw new InvalidParameterError(\n        `dModel (${dModel}) must be divisible by nHead (${nHead})`,\n        \"dModel\",\n        dModel\n      );\n    }\n    if (!Number.isInteger(dFF) || dFF <= 0) {\n      throw new InvalidParameterError(\"dFF must be a positive integer\", \"dFF\", dFF);\n    }\n\n    const dropout = options.dropout ?? 0.1;\n    const eps = options.eps ?? 1e-5;\n\n    this.dModel = dModel;\n    this.nHead = nHead;\n    this.dFF = dFF;\n    this.dropout = dropout;\n\n    this.selfAttn = new MultiheadAttention(dModel, nHead, { dropout });\n    this.linear1 = new Linear(dModel, dFF);\n    this.linear2 = new Linear(dFF, dModel);\n    this.norm1 = new LayerNorm(dModel, { eps });\n    this.norm2 = new LayerNorm(dModel, { eps });\n    this.dropout1 = new Dropout(dropout);\n    this.dropout2 = new Dropout(dropout);\n    this.dropout3 = new Dropout(dropout);\n\n    this.registerModule(\"self_attn\", this.selfAttn);\n    this.registerModule(\"linear1\", this.linear1);\n    this.registerModule(\"linear2\", this.linear2);\n    this.registerModule(\"norm1\", this.norm1);\n    this.registerModule(\"norm2\", this.norm2);\n    this.registerModule(\"dropout1\", this.dropout1);\n    this.registerModule(\"dropout2\", this.dropout2);\n    this.registerModule(\"dropout3\", this.dropout3);\n  }\n\n  /**\n   * Forward pass of the Transformer encoder layer.\n   *\n   * @param src - Source sequence of shape (batch, seqLen, dModel)\n   * @returns Output of same shape as input\n   */\n  forward(src: AnyTensor): GradTensor {\n    const input = src instanceof GradTensor ? src : GradTensor.fromTensor(src);\n\n    if (input.dtype === \"string\") {\n      throw new DTypeError(\"TransformerEncoderLayer does not support string dtype\");\n    }\n\n    // 1. Self Attention\n    // src2 = self_attn(src, src, src)\n    let src2 = this.selfAttn.forward(input, input, input);\n\n    // src = src + dropout(src2)\n    src2 = this.dropout1.forward(src2);\n    let out = input.add(src2);\n\n    // src = norm1(src)\n    out = this.norm1.forward(out);\n\n    // 2. Feed Forward\n    // src2 = linear2(dropout(relu(linear1(src))))\n    // We implement FFN manually with modules\n    let ffn = this.linear1.forward(out);\n    ffn = ffn.relu();\n    ffn = this.dropout2.forward(ffn);\n    ffn = this.linear2.forward(ffn);\n\n    // src = src + dropout(src2)\n    ffn = this.dropout3.forward(ffn);\n    out = out.add(ffn);\n\n    // src = norm2(src)\n    out = this.norm2.forward(out);\n\n    return out;\n  }\n\n  override toString(): string {\n    return `TransformerEncoderLayer(d_model=${this.dModel}, nhead=${this.nHead}, dim_feedforward=${this.dFF}, dropout=${this.dropout})`;\n  }\n}\n","import { DTypeError, InvalidParameterError, NotFittedError, ShapeError } from \"../../core\";\nimport { type AnyTensor, GradTensor, im2colGrad, mulScalar, parameter, randn } from \"../../ndarray\";\nimport { Module } from \"../module/Module\";\n\nfunction normalizePair(\n  name: string,\n  value: number | [number, number],\n  allowZero: boolean,\n  description: string\n): [number, number] {\n  const arr = typeof value === \"number\" ? [value, value] : value;\n  const first = arr[0];\n  const second = arr[1];\n  if (\n    arr.length !== 2 ||\n    first === undefined ||\n    second === undefined ||\n    !Number.isInteger(first) ||\n    !Number.isInteger(second) ||\n    (allowZero ? first < 0 || second < 0 : first <= 0 || second <= 0)\n  ) {\n    throw new InvalidParameterError(`${name} must be ${description}`, name, value);\n  }\n  return [first, second];\n}\n\n/**\n * 1D Convolutional Layer.\n *\n * Applies a 1D convolution over an input signal composed of several input planes.\n *\n * @example\n * ```ts\n * import { Conv1d } from 'deepbox/nn';\n *\n * const conv = new Conv1d(16, 33, 3); // in_channels=16, out_channels=33, kernel_size=3\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.Conv1d.html | PyTorch Conv1d}\n */\nexport class Conv1d extends Module {\n  private readonly inChannels: number;\n  private readonly outChannels: number;\n  private readonly kernelSize: number;\n  private readonly stride: number;\n  private readonly padding: number;\n  private readonly bias: boolean;\n\n  private weight_?: GradTensor;\n  private bias_?: GradTensor;\n\n  constructor(\n    inChannels: number,\n    outChannels: number,\n    kernelSize: number,\n    options: {\n      readonly stride?: number;\n      readonly padding?: number;\n      readonly bias?: boolean;\n    } = {}\n  ) {\n    super();\n\n    // Validate parameters\n    if (inChannels <= 0 || !Number.isInteger(inChannels)) {\n      throw new InvalidParameterError(\n        \"inChannels must be a positive integer\",\n        \"inChannels\",\n        inChannels\n      );\n    }\n    if (outChannels <= 0 || !Number.isInteger(outChannels)) {\n      throw new InvalidParameterError(\n        \"outChannels must be a positive integer\",\n        \"outChannels\",\n        outChannels\n      );\n    }\n    if (kernelSize <= 0 || !Number.isInteger(kernelSize)) {\n      throw new InvalidParameterError(\n        \"kernelSize must be a positive integer\",\n        \"kernelSize\",\n        kernelSize\n      );\n    }\n\n    const stride = options.stride ?? 1;\n    if (stride <= 0 || !Number.isInteger(stride)) {\n      throw new InvalidParameterError(\"stride must be a positive integer\", \"stride\", stride);\n    }\n\n    const padding = options.padding ?? 0;\n    if (padding < 0 || !Number.isInteger(padding)) {\n      throw new InvalidParameterError(\"padding must be a non-negative integer\", \"padding\", padding);\n    }\n\n    this.inChannels = inChannels;\n    this.outChannels = outChannels;\n    this.kernelSize = kernelSize;\n    this.stride = stride;\n    this.padding = padding;\n    this.bias = options.bias ?? true;\n\n    this.initializeParameters();\n  }\n\n  private initializeParameters(): void {\n    const k = 1 / Math.sqrt(this.inChannels * this.kernelSize);\n    const weight = randn([this.outChannels, this.inChannels, this.kernelSize]);\n    this.weight_ = parameter(mulScalar(weight, k));\n    this.registerParameter(\"weight\", this.weight_);\n\n    if (this.bias) {\n      const biasInit = randn([this.outChannels]);\n      this.bias_ = parameter(mulScalar(biasInit, k));\n      this.registerParameter(\"bias\", this.bias_);\n    }\n  }\n\n  forward(x: AnyTensor): GradTensor {\n    // Convert to GradTensor if needed\n    const input = x instanceof GradTensor ? x : GradTensor.fromTensor(x);\n\n    // Reject string tensors\n    if (input.dtype === \"string\") {\n      throw new DTypeError(\"String tensors are not supported\");\n    }\n\n    // Input shape: (batch, in_channels, length)\n    if (input.ndim !== 3) {\n      throw new ShapeError(`Conv1d expects 3D input (batch, channels, length), got ${input.ndim}D`);\n    }\n\n    const batch = input.shape[0] ?? 0;\n    const inC = input.shape[1] ?? 0;\n    const inL = input.shape[2] ?? 0;\n\n    if (inC !== this.inChannels) {\n      throw new ShapeError(`Expected ${this.inChannels} input channels, got ${inC}`);\n    }\n\n    const weight = this.weight_;\n    if (!weight) throw new NotFittedError(\"Weight not initialized\");\n\n    // 1D Convolution using 2D operations (unsqueeze height dim)\n    // Input: (B, C, L) -> (B, C, 1, L)\n    const input2d = input.reshape([batch, inC, 1, inL]);\n\n    // Params for im2col\n    // Kernel: (1, K)\n    const kernelSize: [number, number] = [1, this.kernelSize];\n    const stride: [number, number] = [1, this.stride];\n    const padding: [number, number] = [0, this.padding];\n\n    // im2col -> (B, outL, C * 1 * K)\n    const cols = im2colGrad(input2d, kernelSize, stride, padding);\n\n    // Weights: (outC, inC, K) -> (outC, inC * K)\n    // Note: im2col flattens as channels * kH * kW.\n    // Our weights are (outC, inC, K). Reshape to (outC, inC * K).\n    const weightFlat = weight.reshape([this.outChannels, this.inChannels * this.kernelSize]);\n\n    // Matmul: (B, outL, inC*K) @ (outC, inC*K).T -> (B, outL, outC)\n    const out = cols.matmul(weightFlat.transpose());\n\n    // Reshape to (B, outC, outL)\n    const outTransposed = out.transpose([0, 2, 1]); // (B, outC, outL)\n\n    if (this.bias && this.bias_) {\n      // Bias: (outC) -> (1, outC, 1) broadcast\n      const biasReshaped = this.bias_.reshape([1, this.outChannels, 1]);\n      return outTransposed.add(biasReshaped);\n    }\n\n    return outTransposed;\n  }\n\n  get weight(): GradTensor {\n    if (!this.weight_) {\n      throw new NotFittedError(\"Weight not initialized\");\n    }\n    return this.weight_;\n  }\n}\n\n/**\n * 2D Convolutional Layer.\n *\n * Applies a 2D convolution over an input signal composed of several input planes.\n *\n * @example\n * ```ts\n * import { Conv2d } from 'deepbox/nn';\n *\n * const conv = new Conv2d(3, 64, 3); // RGB to 64 channels, 3x3 kernel\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html | PyTorch Conv2d}\n */\nexport class Conv2d extends Module {\n  private readonly inChannels: number;\n  private readonly outChannels: number;\n  private readonly kernelSize: [number, number];\n  private readonly stride: [number, number];\n  private readonly padding: [number, number];\n  private readonly useBias: boolean;\n\n  private weight_?: GradTensor;\n  private bias_?: GradTensor;\n\n  constructor(\n    inChannels: number,\n    outChannels: number,\n    kernelSize: number | [number, number],\n    options: {\n      readonly stride?: number | [number, number];\n      readonly padding?: number | [number, number];\n      readonly bias?: boolean;\n    } = {}\n  ) {\n    super();\n    if (inChannels <= 0 || !Number.isInteger(inChannels)) {\n      throw new InvalidParameterError(\n        \"inChannels must be a positive integer\",\n        \"inChannels\",\n        inChannels\n      );\n    }\n    if (outChannels <= 0 || !Number.isInteger(outChannels)) {\n      throw new InvalidParameterError(\n        \"outChannels must be a positive integer\",\n        \"outChannels\",\n        outChannels\n      );\n    }\n    const kernelArr = normalizePair(\n      \"kernelSize\",\n      kernelSize,\n      false,\n      \"a positive integer or a tuple of two positive integers\"\n    );\n\n    const stride = options.stride ?? 1;\n    const strideArr = normalizePair(\n      \"stride\",\n      stride,\n      false,\n      \"a positive integer or a tuple of two positive integers\"\n    );\n\n    const padding = options.padding ?? 0;\n    const paddingArr = normalizePair(\n      \"padding\",\n      padding,\n      true,\n      \"a non-negative integer or a tuple of two non-negative integers\"\n    );\n\n    this.inChannels = inChannels;\n    this.outChannels = outChannels;\n    this.kernelSize = kernelArr;\n    this.stride = strideArr;\n    this.padding = paddingArr;\n\n    this.useBias = options.bias ?? true;\n\n    this.initializeParameters();\n  }\n\n  private initializeParameters(): void {\n    const kH = this.kernelSize[0] ?? 1;\n    const kW = this.kernelSize[1] ?? 1;\n    const k = 1 / Math.sqrt(this.inChannels * kH * kW);\n    const weight = randn([this.outChannels, this.inChannels, kH, kW]);\n    this.weight_ = parameter(mulScalar(weight, k));\n    this.registerParameter(\"weight\", this.weight_);\n\n    if (this.useBias) {\n      const biasInit = randn([this.outChannels]);\n      this.bias_ = parameter(mulScalar(biasInit, k));\n      this.registerParameter(\"bias\", this.bias_);\n    }\n  }\n\n  forward(x: AnyTensor): GradTensor {\n    const input = x instanceof GradTensor ? x : GradTensor.fromTensor(x);\n\n    if (input.dtype === \"string\") {\n      throw new DTypeError(\"String tensors are not supported\");\n    }\n\n    if (input.ndim !== 4) {\n      throw new ShapeError(\n        `Conv2d expects 4D input (batch, channels, height, width), got ${input.ndim}D`\n      );\n    }\n\n    const batch = input.shape[0] ?? 0;\n    const inC = input.shape[1] ?? 0;\n    const inH = input.shape[2] ?? 0;\n    const inW = input.shape[3] ?? 0;\n\n    if (inC !== this.inChannels) {\n      throw new ShapeError(`Expected ${this.inChannels} input channels, got ${inC}`);\n    }\n\n    const weight = this.weight_;\n    if (!weight) throw new NotFittedError(\"Weight not initialized\");\n\n    const [kH, kW] = this.kernelSize;\n    const [sH, sW] = this.stride;\n    const [pH, pW] = this.padding;\n\n    // im2col -> (B, outH*outW, C*kH*kW)\n    const cols = im2colGrad(input, [kH, kW], [sH, sW], [pH, pW]);\n\n    // Calculate output dimensions from input (im2col does this internally but we need it for reshape)\n    const outH = Math.floor((inH + 2 * pH - kH) / sH) + 1;\n    const outW = Math.floor((inW + 2 * pW - kW) / sW) + 1;\n\n    // Weights: (outC, inC, kH, kW) -> (outC, inC*kH*kW)\n    const weightFlat = weight.reshape([this.outChannels, this.inChannels * kH * kW]);\n\n    // Matmul: (B, outPixels, inFeatures) @ (outC, inFeatures).T -> (B, outPixels, outC)\n    const out = cols.matmul(weightFlat.transpose());\n\n    // Reshape: (B, outPixels, outC) -> (B, outC, outPixels) -> (B, outC, outH, outW)\n    const outTransposed = out.transpose([0, 2, 1]);\n    const outReshaped = outTransposed.reshape([batch, this.outChannels, outH, outW]);\n\n    if (this.useBias && this.bias_) {\n      // Bias: (outC) -> (1, outC, 1, 1)\n      const biasReshaped = this.bias_.reshape([1, this.outChannels, 1, 1]);\n      return outReshaped.add(biasReshaped);\n    }\n\n    return outReshaped;\n  }\n\n  get weight(): GradTensor {\n    if (!this.weight_) {\n      throw new NotFittedError(\"Weight not initialized\");\n    }\n    return this.weight_;\n  }\n}\n\n/**\n * 2D Max Pooling Layer.\n *\n * Applies a 2D max pooling over an input signal.\n *\n * @example\n * ```ts\n * import { MaxPool2d } from 'deepbox/nn';\n *\n * const pool = new MaxPool2d(2); // 2x2 pooling\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html | PyTorch MaxPool2d}\n */\nexport class MaxPool2d extends Module {\n  private readonly kernelSizeValue: [number, number];\n  private readonly stride: [number, number];\n  private readonly padding: [number, number];\n\n  constructor(\n    kernelSize: number | [number, number],\n    options: {\n      readonly stride?: number | [number, number];\n      readonly padding?: number | [number, number];\n    } = {}\n  ) {\n    super();\n\n    const kernelArr = normalizePair(\n      \"kernelSize\",\n      kernelSize,\n      false,\n      \"a positive integer or a tuple of two positive integers\"\n    );\n    this.kernelSizeValue = kernelArr;\n\n    const strideArr = normalizePair(\n      \"stride\",\n      options.stride ?? kernelSize,\n      false,\n      \"a positive integer or a tuple of two positive integers\"\n    );\n    this.stride = strideArr;\n\n    const paddingArr = normalizePair(\n      \"padding\",\n      options.padding ?? 0,\n      true,\n      \"a non-negative integer or a tuple of two non-negative integers\"\n    );\n    this.padding = paddingArr;\n  }\n\n  forward(x: AnyTensor): GradTensor {\n    const input = x instanceof GradTensor ? x : GradTensor.fromTensor(x);\n\n    if (input.dtype === \"string\") {\n      throw new DTypeError(\"String tensors are not supported\");\n    }\n\n    if (input.ndim !== 4) {\n      throw new ShapeError(\n        `MaxPool2d expects 4D input (batch, channels, height, width), got ${input.ndim}D`\n      );\n    }\n\n    const batch = input.shape[0] ?? 0;\n    const channels = input.shape[1] ?? 0;\n    const inH = input.shape[2] ?? 0;\n    const inW = input.shape[3] ?? 0;\n\n    const [kH, kW] = this.kernelSizeValue;\n    const [sH, sW] = this.stride;\n    const [pH, pW] = this.padding;\n\n    // We use im2col for pooling.\n    // To handle channels correctly (independent pooling per channel),\n    // we treat (batch * channels) as the batch dimension and 1 as channel dimension.\n    // Reshape: (B, C, H, W) -> (B*C, 1, H, W)\n    const inputReshaped = input.reshape([batch * channels, 1, inH, inW]);\n\n    // im2col -> (B*C, outPixels, 1 * kH * kW)\n    const cols = im2colGrad(inputReshaped, [kH, kW], [sH, sW], [pH, pW]);\n\n    // Max over kernel window (axis 2)\n    // (B*C, outPixels, kH*kW) -> (B*C, outPixels)\n    const maxVals = cols.max(2);\n\n    // Calculate output dims\n    const outH = Math.floor((inH + 2 * pH - kH) / sH) + 1;\n    const outW = Math.floor((inW + 2 * pW - kW) / sW) + 1;\n\n    // Reshape back: (B*C, outH*outW) -> (B, C, outH, outW)\n    return maxVals.reshape([batch, channels, outH, outW]);\n  }\n}\n\n/**\n * 2D Average Pooling Layer.\n *\n * Applies a 2D average pooling over an input signal.\n *\n * @example\n * ```ts\n * import { AvgPool2d } from 'deepbox/nn';\n *\n * const pool = new AvgPool2d(2); // 2x2 pooling\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.AvgPool2d.html | PyTorch AvgPool2d}\n */\nexport class AvgPool2d extends Module {\n  private readonly kernelSizeValue: [number, number];\n  private readonly stride: [number, number];\n  private readonly padding: [number, number];\n\n  constructor(\n    kernelSize: number | [number, number],\n    options: {\n      readonly stride?: number | [number, number];\n      readonly padding?: number | [number, number];\n    } = {}\n  ) {\n    super();\n\n    const kernelArr = normalizePair(\n      \"kernelSize\",\n      kernelSize,\n      false,\n      \"a positive integer or a tuple of two positive integers\"\n    );\n    this.kernelSizeValue = kernelArr;\n\n    const strideArr = normalizePair(\n      \"stride\",\n      options.stride ?? kernelSize,\n      false,\n      \"a positive integer or a tuple of two positive integers\"\n    );\n    this.stride = strideArr;\n\n    const paddingArr = normalizePair(\n      \"padding\",\n      options.padding ?? 0,\n      true,\n      \"a non-negative integer or a tuple of two non-negative integers\"\n    );\n    this.padding = paddingArr;\n  }\n\n  forward(x: AnyTensor): GradTensor {\n    const input = x instanceof GradTensor ? x : GradTensor.fromTensor(x);\n\n    if (input.dtype === \"string\") {\n      throw new DTypeError(\"String tensors are not supported\");\n    }\n\n    if (input.ndim !== 4) {\n      throw new ShapeError(\n        `AvgPool2d expects 4D input (batch, channels, height, width), got ${input.ndim}D`\n      );\n    }\n\n    const batch = input.shape[0] ?? 0;\n    const channels = input.shape[1] ?? 0;\n    const inH = input.shape[2] ?? 0;\n    const inW = input.shape[3] ?? 0;\n\n    const [kH, kW] = this.kernelSizeValue;\n    const [sH, sW] = this.stride;\n    const [pH, pW] = this.padding;\n\n    // Reshape: (B, C, H, W) -> (B*C, 1, H, W)\n    const inputReshaped = input.reshape([batch * channels, 1, inH, inW]);\n\n    // im2col -> (B*C, outPixels, 1 * kH * kW)\n    const cols = im2colGrad(inputReshaped, [kH, kW], [sH, sW], [pH, pW]);\n\n    // Mean over kernel window (axis 2)\n    // (B*C, outPixels, kH*kW) -> (B*C, outPixels)\n    const meanVals = cols.mean(2);\n\n    // Calculate output dims\n    const outH = Math.floor((inH + 2 * pH - kH) / sH) + 1;\n    const outW = Math.floor((inW + 2 * pW - kW) / sW) + 1;\n\n    // Reshape back: (B*C, outH*outW) -> (B, C, outH, outW)\n    return meanVals.reshape([batch, channels, outH, outW]);\n  }\n}\n","import {\n  DTypeError,\n  getElementAsNumber,\n  InvalidParameterError,\n  type Shape,\n  ShapeError,\n} from \"../../core\";\nimport {\n  type AnyTensor,\n  GradTensor,\n  mulScalar,\n  parameter,\n  randn,\n  Tensor,\n  zeros,\n} from \"../../ndarray\";\nimport { Module } from \"../module/Module\";\n\nfunction ensureFloatTensor(\n  t: Tensor,\n  context: string\n): asserts t is Tensor<Shape, \"float32\" | \"float64\"> {\n  if (t.dtype === \"string\") {\n    throw new DTypeError(`${context} does not support string dtype`);\n  }\n  if (t.dtype !== \"float32\" && t.dtype !== \"float64\") {\n    throw new DTypeError(`${context} expects float32 or float64 dtype`);\n  }\n}\n\nfunction readNumeric(t: Tensor, offset: number): number {\n  const data = t.data;\n  if (Array.isArray(data)) {\n    throw new DTypeError(\"String tensors are not supported\");\n  }\n  return getElementAsNumber(data, offset);\n}\n\nfunction createFloatBuffer(\n  size: number,\n  dtype: \"float32\" | \"float64\"\n): Float32Array | Float64Array {\n  return dtype === \"float64\" ? new Float64Array(size) : new Float32Array(size);\n}\n\nfunction validatePositiveInt(name: string, value: number): void {\n  if (!Number.isInteger(value) || value <= 0) {\n    throw new InvalidParameterError(`${name} must be a positive integer`, name, value);\n  }\n}\n\ntype ParsedInput = {\n  readonly batch: number;\n  readonly seqLen: number;\n  readonly inputDim: number;\n  readonly isUnbatched: boolean;\n  readonly batchStride: number;\n  readonly seqStride: number;\n  readonly featStride: number;\n};\n\nfunction parseInput(input: Tensor, batchFirst: boolean): ParsedInput {\n  if (input.ndim === 2) {\n    const seqLen = input.shape[0] ?? 0;\n    const inputDim = input.shape[1] ?? 0;\n    return {\n      batch: 1,\n      seqLen,\n      inputDim,\n      isUnbatched: true,\n      batchStride: 0,\n      seqStride: input.strides[0] ?? 0,\n      featStride: input.strides[1] ?? 0,\n    };\n  }\n\n  if (input.ndim !== 3) {\n    throw new ShapeError(`Recurrent layers expect 2D or 3D input; got ndim=${input.ndim}`);\n  }\n\n  if (batchFirst) {\n    return {\n      batch: input.shape[0] ?? 0,\n      seqLen: input.shape[1] ?? 0,\n      inputDim: input.shape[2] ?? 0,\n      isUnbatched: false,\n      batchStride: input.strides[0] ?? 0,\n      seqStride: input.strides[1] ?? 0,\n      featStride: input.strides[2] ?? 0,\n    };\n  }\n\n  return {\n    batch: input.shape[1] ?? 0,\n    seqLen: input.shape[0] ?? 0,\n    inputDim: input.shape[2] ?? 0,\n    isUnbatched: false,\n    batchStride: input.strides[1] ?? 0,\n    seqStride: input.strides[0] ?? 0,\n    featStride: input.strides[2] ?? 0,\n  };\n}\n\nfunction outputIndex(\n  batchFirst: boolean,\n  isUnbatched: boolean,\n  batch: number,\n  seqLen: number,\n  hiddenSize: number,\n  b: number,\n  t: number,\n  j: number\n): number {\n  if (isUnbatched) {\n    return t * hiddenSize + j;\n  }\n  if (batchFirst) {\n    return b * (seqLen * hiddenSize) + t * hiddenSize + j;\n  }\n  return t * (batch * hiddenSize) + b * hiddenSize + j;\n}\n\nfunction extractTensor(arg: AnyTensor, _name: string): Tensor {\n  if (arg instanceof GradTensor) {\n    return arg.tensor;\n  }\n  return arg;\n}\n\nfunction buildState(\n  state: Tensor | undefined,\n  numLayers: number,\n  batch: number,\n  hiddenSize: number,\n  isUnbatched: boolean,\n  name: string\n): Float64Array[] {\n  const result = new Array<Float64Array>(numLayers);\n  for (let l = 0; l < numLayers; l++) {\n    result[l] = new Float64Array(batch * hiddenSize);\n  }\n\n  if (!state) {\n    return result;\n  }\n\n  ensureFloatTensor(state, name);\n\n  if (state.ndim === 2) {\n    if (!isUnbatched) {\n      throw new ShapeError(`Expected ${name} with 3 dimensions for batched input`);\n    }\n    if ((state.shape[0] ?? 0) !== numLayers || (state.shape[1] ?? 0) !== hiddenSize) {\n      throw new ShapeError(\n        `Expected ${name} shape [${numLayers}, ${hiddenSize}], got [${state.shape.join(\", \")}]`\n      );\n    }\n\n    const stride0 = state.strides[0] ?? 0;\n    const stride1 = state.strides[1] ?? 0;\n    for (let l = 0; l < numLayers; l++) {\n      const layerState = result[l];\n      if (!layerState) {\n        throw new ShapeError(`Internal error: missing ${name} layer state`);\n      }\n      const base = state.offset + l * stride0;\n      for (let j = 0; j < hiddenSize; j++) {\n        layerState[j] = readNumeric(state, base + j * stride1);\n      }\n    }\n\n    return result;\n  }\n\n  if (state.ndim !== 3) {\n    throw new ShapeError(`Expected ${name} with 2 or 3 dimensions; got ndim=${state.ndim}`);\n  }\n\n  const expectedBatch = isUnbatched ? 1 : batch;\n  if (\n    (state.shape[0] ?? 0) !== numLayers ||\n    (state.shape[1] ?? 0) !== expectedBatch ||\n    (state.shape[2] ?? 0) !== hiddenSize\n  ) {\n    const expected = isUnbatched ? [numLayers, 1, hiddenSize] : [numLayers, batch, hiddenSize];\n    throw new ShapeError(\n      `Expected ${name} shape [${expected.join(\", \")}], got [${state.shape.join(\", \")}]`\n    );\n  }\n\n  const stride0 = state.strides[0] ?? 0;\n  const stride1 = state.strides[1] ?? 0;\n  const stride2 = state.strides[2] ?? 0;\n\n  for (let l = 0; l < numLayers; l++) {\n    const layerState = result[l];\n    if (!layerState) {\n      throw new ShapeError(`Internal error: missing ${name} layer state`);\n    }\n    const baseLayer = state.offset + l * stride0;\n    for (let b = 0; b < batch; b++) {\n      const baseBatch = baseLayer + b * stride1;\n      for (let j = 0; j < hiddenSize; j++) {\n        layerState[b * hiddenSize + j] = readNumeric(state, baseBatch + j * stride2);\n      }\n    }\n  }\n\n  return result;\n}\n\nfunction packState(\n  state: Float64Array[],\n  numLayers: number,\n  batch: number,\n  hiddenSize: number,\n  dtype: \"float32\" | \"float64\",\n  device: Tensor[\"device\"],\n  isUnbatched: boolean\n): Tensor {\n  const size = isUnbatched ? numLayers * hiddenSize : numLayers * batch * hiddenSize;\n  const data = createFloatBuffer(size, dtype);\n\n  if (isUnbatched) {\n    for (let l = 0; l < numLayers; l++) {\n      const layer = state[l];\n      if (!layer) {\n        throw new ShapeError(\"Internal error: missing packed state layer\");\n      }\n      for (let j = 0; j < hiddenSize; j++) {\n        data[l * hiddenSize + j] = layer[j] ?? 0;\n      }\n    }\n    return Tensor.fromTypedArray({\n      data,\n      shape: [numLayers, hiddenSize],\n      dtype,\n      device,\n    });\n  }\n\n  for (let l = 0; l < numLayers; l++) {\n    const layer = state[l];\n    if (!layer) {\n      throw new ShapeError(\"Internal error: missing packed state layer\");\n    }\n    const layerOffset = l * batch * hiddenSize;\n    for (let b = 0; b < batch; b++) {\n      const batchOffset = layerOffset + b * hiddenSize;\n      for (let j = 0; j < hiddenSize; j++) {\n        data[batchOffset + j] = layer[b * hiddenSize + j] ?? 0;\n      }\n    }\n  }\n\n  return Tensor.fromTypedArray({\n    data,\n    shape: [numLayers, batch, hiddenSize],\n    dtype,\n    device,\n  });\n}\n\n/**\n * Simple RNN layer.\n *\n * Applies a simple recurrent neural network to an input sequence.\n *\n * **Formula**: h_t = tanh(W_ih * x_t + b_ih + W_hh * h_{t-1} + b_hh)\n *\n * @example\n * ```ts\n * import { RNN } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const rnn = new RNN(10, 20);\n * const x = tensor([[[1, 2, 3]]]);  // (batch, seq_len, input_size)\n * const output = rnn.forward(x);\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.RNN.html | PyTorch RNN}\n */\nexport class RNN extends Module {\n  private readonly inputSize: number;\n  private readonly hiddenSize: number;\n  private readonly numLayers: number;\n  private readonly nonlinearity: \"tanh\" | \"relu\";\n  private readonly bias: boolean;\n  private readonly batchFirst: boolean;\n\n  private weightsIh: Tensor[];\n  private weightsHh: Tensor[];\n  private biasIh: Tensor[];\n  private biasHh: Tensor[];\n\n  constructor(\n    inputSize: number,\n    hiddenSize: number,\n    options: {\n      readonly numLayers?: number;\n      readonly nonlinearity?: \"tanh\" | \"relu\";\n      readonly bias?: boolean;\n      readonly batchFirst?: boolean;\n    } = {}\n  ) {\n    super();\n    validatePositiveInt(\"inputSize\", inputSize);\n    validatePositiveInt(\"hiddenSize\", hiddenSize);\n    const numLayers = options.numLayers ?? 1;\n    validatePositiveInt(\"numLayers\", numLayers);\n\n    this.inputSize = inputSize;\n    this.hiddenSize = hiddenSize;\n    this.numLayers = numLayers;\n    this.nonlinearity = options.nonlinearity ?? \"tanh\";\n    this.bias = options.bias ?? true;\n    this.batchFirst = options.batchFirst ?? true;\n\n    const stdv = 1.0 / Math.sqrt(hiddenSize);\n\n    this.weightsIh = [];\n    this.weightsHh = [];\n    this.biasIh = [];\n    this.biasHh = [];\n\n    for (let layer = 0; layer < this.numLayers; layer++) {\n      const inputDim = layer === 0 ? inputSize : hiddenSize;\n      const wIh = mulScalar(randn([hiddenSize, inputDim]), stdv);\n      const wHh = mulScalar(randn([hiddenSize, hiddenSize]), stdv);\n\n      this.weightsIh.push(wIh);\n      this.weightsHh.push(wHh);\n      this.registerParameter(`weight_ih_l${layer}`, parameter(wIh));\n      this.registerParameter(`weight_hh_l${layer}`, parameter(wHh));\n\n      if (this.bias) {\n        const bIh = zeros([hiddenSize]);\n        const bHh = zeros([hiddenSize]);\n        this.biasIh.push(bIh);\n        this.biasHh.push(bHh);\n        this.registerParameter(`bias_ih_l${layer}`, parameter(bIh));\n        this.registerParameter(`bias_hh_l${layer}`, parameter(bHh));\n      }\n    }\n  }\n\n  private activation(x: number): number {\n    return this.nonlinearity === \"tanh\" ? Math.tanh(x) : Math.max(0, x);\n  }\n\n  private run(input: Tensor, hx?: Tensor): { output: Tensor; h: Tensor } {\n    ensureFloatTensor(input, \"RNN\");\n    const parsed = parseInput(input, this.batchFirst);\n    const { batch, seqLen, inputDim, isUnbatched, batchStride, seqStride, featStride } = parsed;\n\n    if (inputDim !== this.inputSize) {\n      throw new ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);\n    }\n    if (seqLen <= 0) {\n      throw new InvalidParameterError(\"Sequence length must be positive\", \"seqLen\", seqLen);\n    }\n    if (!isUnbatched && batch <= 0) {\n      throw new InvalidParameterError(\"Batch size must be positive\", \"batch\", batch);\n    }\n\n    const h = buildState(hx, this.numLayers, batch, this.hiddenSize, isUnbatched, \"hx\");\n    const outSize = (isUnbatched ? seqLen : batch * seqLen) * this.hiddenSize;\n    const out = createFloatBuffer(outSize, input.dtype);\n\n    const inputVec = new Float64Array(inputDim);\n\n    for (let t = 0; t < seqLen; t++) {\n      for (let b = 0; b < batch; b++) {\n        const baseOffset = input.offset + b * batchStride + t * seqStride;\n        for (let i = 0; i < inputDim; i++) {\n          inputVec[i] = readNumeric(input, baseOffset + i * featStride);\n        }\n\n        let layerInput = inputVec;\n\n        for (let l = 0; l < this.numLayers; l++) {\n          const wIh = this.weightsIh[l];\n          const wHh = this.weightsHh[l];\n          if (!wIh || !wHh) {\n            throw new ShapeError(\"Internal error: missing RNN weights\");\n          }\n\n          const curInputSize = l === 0 ? this.inputSize : this.hiddenSize;\n          const newH = new Float64Array(this.hiddenSize);\n          const hLayer = h[l];\n          if (!hLayer) {\n            throw new ShapeError(\"Internal error: missing RNN hidden state\");\n          }\n\n          const wIhStride0 = wIh.strides[0] ?? 0;\n          const wIhStride1 = wIh.strides[1] ?? 0;\n          const wHhStride0 = wHh.strides[0] ?? 0;\n          const wHhStride1 = wHh.strides[1] ?? 0;\n          const biasIh = this.biasIh[l];\n          const biasHh = this.biasHh[l];\n          const biasIhStride = biasIh ? (biasIh.strides[0] ?? 0) : 0;\n          const biasHhStride = biasHh ? (biasHh.strides[0] ?? 0) : 0;\n\n          for (let j = 0; j < this.hiddenSize; j++) {\n            let sum = 0;\n            const wIhBase = wIh.offset + j * wIhStride0;\n            for (let k = 0; k < curInputSize; k++) {\n              sum += (layerInput[k] ?? 0) * readNumeric(wIh, wIhBase + k * wIhStride1);\n            }\n\n            const wHhBase = wHh.offset + j * wHhStride0;\n            for (let k = 0; k < this.hiddenSize; k++) {\n              sum +=\n                (hLayer[b * this.hiddenSize + k] ?? 0) * readNumeric(wHh, wHhBase + k * wHhStride1);\n            }\n\n            if (this.bias && biasIh && biasHh) {\n              sum += readNumeric(biasIh, biasIh.offset + j * biasIhStride);\n              sum += readNumeric(biasHh, biasHh.offset + j * biasHhStride);\n            }\n\n            newH[j] = this.activation(sum);\n          }\n\n          for (let j = 0; j < this.hiddenSize; j++) {\n            hLayer[b * this.hiddenSize + j] = newH[j] ?? 0;\n          }\n\n          layerInput = newH;\n        }\n\n        for (let j = 0; j < this.hiddenSize; j++) {\n          const idx = outputIndex(\n            this.batchFirst,\n            isUnbatched,\n            batch,\n            seqLen,\n            this.hiddenSize,\n            b,\n            t,\n            j\n          );\n          out[idx] = layerInput[j] ?? 0;\n        }\n      }\n    }\n\n    const outShape = isUnbatched\n      ? [seqLen, this.hiddenSize]\n      : this.batchFirst\n        ? [batch, seqLen, this.hiddenSize]\n        : [seqLen, batch, this.hiddenSize];\n\n    return {\n      output: Tensor.fromTypedArray({\n        data: out,\n        shape: outShape,\n        dtype: input.dtype,\n        device: input.device,\n      }),\n      h: packState(\n        h,\n        this.numLayers,\n        batch,\n        this.hiddenSize,\n        input.dtype,\n        input.device,\n        isUnbatched\n      ),\n    };\n  }\n\n  forward(...inputs: AnyTensor[]): Tensor {\n    if (inputs.length < 1 || inputs.length > 2) {\n      throw new InvalidParameterError(\"RNN.forward expects 1 or 2 inputs\", \"inputs\", inputs.length);\n    }\n    const inputArg = inputs[0];\n    if (inputArg === undefined) {\n      throw new InvalidParameterError(\"RNN.forward requires an input tensor\", \"input\", inputArg);\n    }\n    const input = extractTensor(inputArg, \"input\");\n    const hxArg = inputs.length === 2 ? inputs[1] : undefined;\n    const hx = hxArg === undefined ? undefined : extractTensor(hxArg, \"hx\");\n    return this.run(input, hx).output;\n  }\n\n  /**\n   * Forward pass returning both output and hidden state.\n   * Use this method when you need the hidden state.\n   */\n  forwardWithState(input: AnyTensor, hx?: AnyTensor): [Tensor, Tensor] {\n    const inputTensor = extractTensor(input, \"input\");\n    const hxTensor = hx === undefined ? undefined : extractTensor(hx, \"hx\");\n    const { output, h } = this.run(inputTensor, hxTensor);\n    return [output, h];\n  }\n\n  override toString(): string {\n    return `RNN(${this.inputSize}, ${this.hiddenSize}, num_layers=${this.numLayers})`;\n  }\n}\n\n/**\n * LSTM (Long Short-Term Memory) layer.\n *\n * Applies a multi-layer LSTM to an input sequence.\n *\n * **Gates**:\n * - Input gate: i_t = σ(W_ii * x_t + b_ii + W_hi * h_{t-1} + b_hi)\n * - Forget gate: f_t = σ(W_if * x_t + b_if + W_hf * h_{t-1} + b_hf)\n * - Cell gate: g_t = tanh(W_ig * x_t + b_ig + W_hg * h_{t-1} + b_hg)\n * - Output gate: o_t = σ(W_io * x_t + b_io + W_ho * h_{t-1} + b_ho)\n * - Cell state: c_t = f_t * c_{t-1} + i_t * g_t\n * - Hidden state: h_t = o_t * tanh(c_t)\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.LSTM.html | PyTorch LSTM}\n */\nexport class LSTM extends Module {\n  private readonly inputSize: number;\n  private readonly hiddenSize: number;\n  private readonly numLayers: number;\n  private readonly bias: boolean;\n  private readonly batchFirst: boolean;\n\n  private weightsIh: Tensor[];\n  private weightsHh: Tensor[];\n  private biasIh: Tensor[];\n  private biasHh: Tensor[];\n\n  constructor(\n    inputSize: number,\n    hiddenSize: number,\n    options: {\n      readonly numLayers?: number;\n      readonly bias?: boolean;\n      readonly batchFirst?: boolean;\n    } = {}\n  ) {\n    super();\n    validatePositiveInt(\"inputSize\", inputSize);\n    validatePositiveInt(\"hiddenSize\", hiddenSize);\n    const numLayers = options.numLayers ?? 1;\n    validatePositiveInt(\"numLayers\", numLayers);\n\n    this.inputSize = inputSize;\n    this.hiddenSize = hiddenSize;\n    this.numLayers = numLayers;\n    this.bias = options.bias ?? true;\n    this.batchFirst = options.batchFirst ?? true;\n\n    const stdv = 1.0 / Math.sqrt(hiddenSize);\n\n    this.weightsIh = [];\n    this.weightsHh = [];\n    this.biasIh = [];\n    this.biasHh = [];\n\n    for (let layer = 0; layer < this.numLayers; layer++) {\n      const inputDim = layer === 0 ? inputSize : hiddenSize;\n\n      const wIh = mulScalar(randn([4 * hiddenSize, inputDim]), stdv);\n      const wHh = mulScalar(randn([4 * hiddenSize, hiddenSize]), stdv);\n\n      this.weightsIh.push(wIh);\n      this.weightsHh.push(wHh);\n      this.registerParameter(`weight_ih_l${layer}`, parameter(wIh));\n      this.registerParameter(`weight_hh_l${layer}`, parameter(wHh));\n\n      if (this.bias) {\n        const bIh = zeros([4 * hiddenSize]);\n        const bHh = zeros([4 * hiddenSize]);\n        this.biasIh.push(bIh);\n        this.biasHh.push(bHh);\n        this.registerParameter(`bias_ih_l${layer}`, parameter(bIh));\n        this.registerParameter(`bias_hh_l${layer}`, parameter(bHh));\n      }\n    }\n  }\n\n  private sigmoid(x: number): number {\n    return 1 / (1 + Math.exp(-x));\n  }\n\n  private run(input: Tensor, hx?: Tensor, cx?: Tensor): { output: Tensor; h: Tensor; c: Tensor } {\n    ensureFloatTensor(input, \"LSTM\");\n    const parsed = parseInput(input, this.batchFirst);\n    const { batch, seqLen, inputDim, isUnbatched, batchStride, seqStride, featStride } = parsed;\n\n    if (inputDim !== this.inputSize) {\n      throw new ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);\n    }\n    if (seqLen <= 0) {\n      throw new InvalidParameterError(\"Sequence length must be positive\", \"seqLen\", seqLen);\n    }\n    if (!isUnbatched && batch <= 0) {\n      throw new InvalidParameterError(\"Batch size must be positive\", \"batch\", batch);\n    }\n\n    const h = buildState(hx, this.numLayers, batch, this.hiddenSize, isUnbatched, \"hx\");\n    const c = buildState(cx, this.numLayers, batch, this.hiddenSize, isUnbatched, \"cx\");\n\n    const outSize = (isUnbatched ? seqLen : batch * seqLen) * this.hiddenSize;\n    const out = createFloatBuffer(outSize, input.dtype);\n\n    const inputVec = new Float64Array(inputDim);\n    const gates = new Float64Array(4 * this.hiddenSize);\n\n    for (let t = 0; t < seqLen; t++) {\n      for (let b = 0; b < batch; b++) {\n        const baseOffset = input.offset + b * batchStride + t * seqStride;\n        for (let i = 0; i < inputDim; i++) {\n          inputVec[i] = readNumeric(input, baseOffset + i * featStride);\n        }\n\n        let layerInput = inputVec;\n\n        for (let l = 0; l < this.numLayers; l++) {\n          const wIh = this.weightsIh[l];\n          const wHh = this.weightsHh[l];\n          if (!wIh || !wHh) {\n            throw new ShapeError(\"Internal error: missing LSTM weights\");\n          }\n\n          const curInputSize = l === 0 ? this.inputSize : this.hiddenSize;\n          const hLayer = h[l];\n          const cLayer = c[l];\n          if (!hLayer || !cLayer) {\n            throw new ShapeError(\"Internal error: missing LSTM state\");\n          }\n\n          const wIhStride0 = wIh.strides[0] ?? 0;\n          const wIhStride1 = wIh.strides[1] ?? 0;\n          const wHhStride0 = wHh.strides[0] ?? 0;\n          const wHhStride1 = wHh.strides[1] ?? 0;\n          const biasIh = this.biasIh[l];\n          const biasHh = this.biasHh[l];\n          const biasIhStride = biasIh ? (biasIh.strides[0] ?? 0) : 0;\n          const biasHhStride = biasHh ? (biasHh.strides[0] ?? 0) : 0;\n\n          for (let g = 0; g < 4 * this.hiddenSize; g++) {\n            let sum = 0;\n            const wIhBase = wIh.offset + g * wIhStride0;\n            for (let k = 0; k < curInputSize; k++) {\n              sum += (layerInput[k] ?? 0) * readNumeric(wIh, wIhBase + k * wIhStride1);\n            }\n            const wHhBase = wHh.offset + g * wHhStride0;\n            for (let k = 0; k < this.hiddenSize; k++) {\n              sum +=\n                (hLayer[b * this.hiddenSize + k] ?? 0) * readNumeric(wHh, wHhBase + k * wHhStride1);\n            }\n            if (this.bias && biasIh && biasHh) {\n              sum += readNumeric(biasIh, biasIh.offset + g * biasIhStride);\n              sum += readNumeric(biasHh, biasHh.offset + g * biasHhStride);\n            }\n            gates[g] = sum;\n          }\n\n          const newH = new Float64Array(this.hiddenSize);\n          const newC = new Float64Array(this.hiddenSize);\n\n          for (let j = 0; j < this.hiddenSize; j++) {\n            const iGate = this.sigmoid(gates[j] ?? 0);\n            const fGate = this.sigmoid(gates[this.hiddenSize + j] ?? 0);\n            const gGate = Math.tanh(gates[2 * this.hiddenSize + j] ?? 0);\n            const oGate = this.sigmoid(gates[3 * this.hiddenSize + j] ?? 0);\n\n            const prevC = cLayer[b * this.hiddenSize + j] ?? 0;\n            const nextC = fGate * prevC + iGate * gGate;\n            const nextH = oGate * Math.tanh(nextC);\n\n            newC[j] = nextC;\n            newH[j] = nextH;\n          }\n\n          for (let j = 0; j < this.hiddenSize; j++) {\n            hLayer[b * this.hiddenSize + j] = newH[j] ?? 0;\n            cLayer[b * this.hiddenSize + j] = newC[j] ?? 0;\n          }\n\n          layerInput = newH;\n        }\n\n        for (let j = 0; j < this.hiddenSize; j++) {\n          const idx = outputIndex(\n            this.batchFirst,\n            isUnbatched,\n            batch,\n            seqLen,\n            this.hiddenSize,\n            b,\n            t,\n            j\n          );\n          out[idx] = layerInput[j] ?? 0;\n        }\n      }\n    }\n\n    const outShape = isUnbatched\n      ? [seqLen, this.hiddenSize]\n      : this.batchFirst\n        ? [batch, seqLen, this.hiddenSize]\n        : [seqLen, batch, this.hiddenSize];\n\n    return {\n      output: Tensor.fromTypedArray({\n        data: out,\n        shape: outShape,\n        dtype: input.dtype,\n        device: input.device,\n      }),\n      h: packState(\n        h,\n        this.numLayers,\n        batch,\n        this.hiddenSize,\n        input.dtype,\n        input.device,\n        isUnbatched\n      ),\n      c: packState(\n        c,\n        this.numLayers,\n        batch,\n        this.hiddenSize,\n        input.dtype,\n        input.device,\n        isUnbatched\n      ),\n    };\n  }\n\n  forward(...inputs: AnyTensor[]): Tensor {\n    if (inputs.length < 1 || inputs.length > 3) {\n      throw new InvalidParameterError(\n        \"LSTM.forward expects 1 to 3 inputs\",\n        \"inputs\",\n        inputs.length\n      );\n    }\n    const inputArg = inputs[0];\n    if (inputArg === undefined) {\n      throw new InvalidParameterError(\"LSTM.forward requires an input tensor\", \"input\", inputArg);\n    }\n    const input = extractTensor(inputArg, \"input\");\n    const hxArg = inputs.length >= 2 ? inputs[1] : undefined;\n    const cxArg = inputs.length >= 3 ? inputs[2] : undefined;\n    const hx = hxArg === undefined ? undefined : extractTensor(hxArg, \"hx\");\n    const cx = cxArg === undefined ? undefined : extractTensor(cxArg, \"cx\");\n    return this.run(input, hx, cx).output;\n  }\n\n  /**\n   * Forward pass returning output, hidden state, and cell state.\n   * Use this method when you need the hidden/cell states.\n   */\n  forwardWithState(input: AnyTensor, hx?: AnyTensor, cx?: AnyTensor): [Tensor, [Tensor, Tensor]] {\n    const inputTensor = extractTensor(input, \"input\");\n    const hxTensor = hx === undefined ? undefined : extractTensor(hx, \"hx\");\n    const cxTensor = cx === undefined ? undefined : extractTensor(cx, \"cx\");\n    const { output, h, c } = this.run(inputTensor, hxTensor, cxTensor);\n    return [output, [h, c]];\n  }\n\n  override toString(): string {\n    return `LSTM(${this.inputSize}, ${this.hiddenSize}, num_layers=${this.numLayers})`;\n  }\n}\n\n/**\n * GRU (Gated Recurrent Unit) layer.\n *\n * Applies a multi-layer GRU to an input sequence.\n *\n * **Gates**:\n * - Reset gate: r_t = σ(W_ir * x_t + b_ir + W_hr * h_{t-1} + b_hr)\n * - Update gate: z_t = σ(W_iz * x_t + b_iz + W_hz * h_{t-1} + b_hz)\n * - New gate: n_t = tanh(W_in * x_t + b_in + r_t * (W_hn * h_{t-1} + b_hn))\n * - Hidden: h_t = (1 - z_t) * n_t + z_t * h_{t-1}\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.nn.GRU.html | PyTorch GRU}\n */\nexport class GRU extends Module {\n  private readonly inputSize: number;\n  private readonly hiddenSize: number;\n  private readonly numLayers: number;\n  private readonly bias: boolean;\n  private readonly batchFirst: boolean;\n\n  private weightsIh: Tensor[];\n  private weightsHh: Tensor[];\n  private biasIh: Tensor[];\n  private biasHh: Tensor[];\n\n  constructor(\n    inputSize: number,\n    hiddenSize: number,\n    options: {\n      readonly numLayers?: number;\n      readonly bias?: boolean;\n      readonly batchFirst?: boolean;\n    } = {}\n  ) {\n    super();\n    validatePositiveInt(\"inputSize\", inputSize);\n    validatePositiveInt(\"hiddenSize\", hiddenSize);\n    const numLayers = options.numLayers ?? 1;\n    validatePositiveInt(\"numLayers\", numLayers);\n\n    this.inputSize = inputSize;\n    this.hiddenSize = hiddenSize;\n    this.numLayers = numLayers;\n    this.bias = options.bias ?? true;\n    this.batchFirst = options.batchFirst ?? true;\n\n    const stdv = 1.0 / Math.sqrt(hiddenSize);\n\n    this.weightsIh = [];\n    this.weightsHh = [];\n    this.biasIh = [];\n    this.biasHh = [];\n\n    for (let layer = 0; layer < this.numLayers; layer++) {\n      const inputDim = layer === 0 ? inputSize : hiddenSize;\n\n      const wIh = mulScalar(randn([3 * hiddenSize, inputDim]), stdv);\n      const wHh = mulScalar(randn([3 * hiddenSize, hiddenSize]), stdv);\n\n      this.weightsIh.push(wIh);\n      this.weightsHh.push(wHh);\n      this.registerParameter(`weight_ih_l${layer}`, parameter(wIh));\n      this.registerParameter(`weight_hh_l${layer}`, parameter(wHh));\n\n      if (this.bias) {\n        const bIh = zeros([3 * hiddenSize]);\n        const bHh = zeros([3 * hiddenSize]);\n        this.biasIh.push(bIh);\n        this.biasHh.push(bHh);\n        this.registerParameter(`bias_ih_l${layer}`, parameter(bIh));\n        this.registerParameter(`bias_hh_l${layer}`, parameter(bHh));\n      }\n    }\n  }\n\n  private sigmoid(x: number): number {\n    return 1 / (1 + Math.exp(-x));\n  }\n\n  private run(input: Tensor, hx?: Tensor): { output: Tensor; h: Tensor } {\n    ensureFloatTensor(input, \"GRU\");\n    const parsed = parseInput(input, this.batchFirst);\n    const { batch, seqLen, inputDim, isUnbatched, batchStride, seqStride, featStride } = parsed;\n\n    if (inputDim !== this.inputSize) {\n      throw new ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);\n    }\n    if (seqLen <= 0) {\n      throw new InvalidParameterError(\"Sequence length must be positive\", \"seqLen\", seqLen);\n    }\n    if (!isUnbatched && batch <= 0) {\n      throw new InvalidParameterError(\"Batch size must be positive\", \"batch\", batch);\n    }\n\n    const h = buildState(hx, this.numLayers, batch, this.hiddenSize, isUnbatched, \"hx\");\n    const outSize = (isUnbatched ? seqLen : batch * seqLen) * this.hiddenSize;\n    const out = createFloatBuffer(outSize, input.dtype);\n\n    const inputVec = new Float64Array(inputDim);\n    const gatesIh = new Float64Array(3 * this.hiddenSize);\n    const gatesHh = new Float64Array(3 * this.hiddenSize);\n\n    for (let t = 0; t < seqLen; t++) {\n      for (let b = 0; b < batch; b++) {\n        const baseOffset = input.offset + b * batchStride + t * seqStride;\n        for (let i = 0; i < inputDim; i++) {\n          inputVec[i] = readNumeric(input, baseOffset + i * featStride);\n        }\n\n        let layerInput = inputVec;\n\n        for (let l = 0; l < this.numLayers; l++) {\n          const wIh = this.weightsIh[l];\n          const wHh = this.weightsHh[l];\n          if (!wIh || !wHh) {\n            throw new ShapeError(\"Internal error: missing GRU weights\");\n          }\n\n          const curInputSize = l === 0 ? this.inputSize : this.hiddenSize;\n          const hLayer = h[l];\n          if (!hLayer) {\n            throw new ShapeError(\"Internal error: missing GRU hidden state\");\n          }\n\n          const wIhStride0 = wIh.strides[0] ?? 0;\n          const wIhStride1 = wIh.strides[1] ?? 0;\n          const wHhStride0 = wHh.strides[0] ?? 0;\n          const wHhStride1 = wHh.strides[1] ?? 0;\n          const biasIh = this.biasIh[l];\n          const biasHh = this.biasHh[l];\n          const biasIhStride = biasIh ? (biasIh.strides[0] ?? 0) : 0;\n          const biasHhStride = biasHh ? (biasHh.strides[0] ?? 0) : 0;\n\n          for (let g = 0; g < 3 * this.hiddenSize; g++) {\n            let sumIh = 0;\n            let sumHh = 0;\n            const wIhBase = wIh.offset + g * wIhStride0;\n            for (let k = 0; k < curInputSize; k++) {\n              sumIh += (layerInput[k] ?? 0) * readNumeric(wIh, wIhBase + k * wIhStride1);\n            }\n            const wHhBase = wHh.offset + g * wHhStride0;\n            for (let k = 0; k < this.hiddenSize; k++) {\n              sumHh +=\n                (hLayer[b * this.hiddenSize + k] ?? 0) * readNumeric(wHh, wHhBase + k * wHhStride1);\n            }\n            if (this.bias && biasIh && biasHh) {\n              sumIh += readNumeric(biasIh, biasIh.offset + g * biasIhStride);\n              sumHh += readNumeric(biasHh, biasHh.offset + g * biasHhStride);\n            }\n            gatesIh[g] = sumIh;\n            gatesHh[g] = sumHh;\n          }\n\n          const newH = new Float64Array(this.hiddenSize);\n          for (let j = 0; j < this.hiddenSize; j++) {\n            const r = this.sigmoid((gatesIh[j] ?? 0) + (gatesHh[j] ?? 0));\n            const z = this.sigmoid(\n              (gatesIh[this.hiddenSize + j] ?? 0) + (gatesHh[this.hiddenSize + j] ?? 0)\n            );\n            const n = Math.tanh(\n              (gatesIh[2 * this.hiddenSize + j] ?? 0) + r * (gatesHh[2 * this.hiddenSize + j] ?? 0)\n            );\n            newH[j] = (1 - z) * n + z * (hLayer[b * this.hiddenSize + j] ?? 0);\n          }\n\n          for (let j = 0; j < this.hiddenSize; j++) {\n            hLayer[b * this.hiddenSize + j] = newH[j] ?? 0;\n          }\n\n          layerInput = newH;\n        }\n\n        for (let j = 0; j < this.hiddenSize; j++) {\n          const idx = outputIndex(\n            this.batchFirst,\n            isUnbatched,\n            batch,\n            seqLen,\n            this.hiddenSize,\n            b,\n            t,\n            j\n          );\n          out[idx] = layerInput[j] ?? 0;\n        }\n      }\n    }\n\n    const outShape = isUnbatched\n      ? [seqLen, this.hiddenSize]\n      : this.batchFirst\n        ? [batch, seqLen, this.hiddenSize]\n        : [seqLen, batch, this.hiddenSize];\n\n    return {\n      output: Tensor.fromTypedArray({\n        data: out,\n        shape: outShape,\n        dtype: input.dtype,\n        device: input.device,\n      }),\n      h: packState(\n        h,\n        this.numLayers,\n        batch,\n        this.hiddenSize,\n        input.dtype,\n        input.device,\n        isUnbatched\n      ),\n    };\n  }\n\n  forward(...inputs: AnyTensor[]): Tensor {\n    if (inputs.length < 1 || inputs.length > 2) {\n      throw new InvalidParameterError(\"GRU.forward expects 1 or 2 inputs\", \"inputs\", inputs.length);\n    }\n    const inputArg = inputs[0];\n    if (inputArg === undefined) {\n      throw new InvalidParameterError(\"GRU.forward requires an input tensor\", \"input\", inputArg);\n    }\n    const input = extractTensor(inputArg, \"input\");\n    const hxArg = inputs.length === 2 ? inputs[1] : undefined;\n    const hx = hxArg === undefined ? undefined : extractTensor(hxArg, \"hx\");\n    return this.run(input, hx).output;\n  }\n\n  /**\n   * Forward pass returning both output and hidden state.\n   * Use this method when you need the hidden state.\n   */\n  forwardWithState(input: AnyTensor, hx?: AnyTensor): [Tensor, Tensor] {\n    const inputTensor = extractTensor(input, \"input\");\n    const hxTensor = hx === undefined ? undefined : extractTensor(hx, \"hx\");\n    const { output, h } = this.run(inputTensor, hxTensor);\n    return [output, h];\n  }\n\n  override toString(): string {\n    return `GRU(${this.inputSize}, ${this.hiddenSize}, num_layers=${this.numLayers})`;\n  }\n}\n","import {\n  DTypeError,\n  getBigIntElement,\n  getNumericElement,\n  InvalidParameterError,\n  ShapeError,\n} from \"../../core\";\nimport { type AnyTensor, GradTensor, logSoftmaxGrad, Tensor } from \"../../ndarray\";\n\n/**\n * Helper to convert class indices to one-hot encoded tensor.\n */\nfunction toOneHot(indices: Tensor, numClasses: number): Tensor {\n  const nSamples = indices.size;\n  const outData = new Float32Array(nSamples * numClasses);\n\n  const data = indices.data;\n  if (Array.isArray(data)) {\n    throw new DTypeError(\"crossEntropyLoss target indices must be numeric\");\n  }\n\n  const stride0 = indices.strides[0] ?? 0;\n  const base = indices.offset;\n\n  for (let i = 0; i < nSamples; i++) {\n    const offset = base + i * stride0;\n    let idx: number;\n    if (data instanceof BigInt64Array) {\n      const raw = getBigIntElement(data, offset);\n      const asNumber = Number(raw);\n      if (!Number.isSafeInteger(asNumber)) {\n        throw new InvalidParameterError(\n          `Class index ${raw.toString()} exceeds safe integer range`,\n          \"target\",\n          raw.toString()\n        );\n      }\n      idx = asNumber;\n    } else {\n      idx = Number(getNumericElement(data, offset));\n    }\n\n    if (!Number.isFinite(idx) || !Number.isInteger(idx)) {\n      throw new InvalidParameterError(`Class index ${idx} is not a valid integer`, \"target\", idx);\n    }\n\n    if (idx < 0 || idx >= numClasses) {\n      throw new InvalidParameterError(\n        `Class index ${idx} out of range [0, ${numClasses})`,\n        \"target\",\n        idx\n      );\n    }\n    outData[i * numClasses + idx] = 1.0;\n  }\n\n  return Tensor.fromTypedArray({\n    data: outData,\n    shape: [nSamples, numClasses],\n    dtype: \"float32\",\n    device: indices.device,\n  });\n}\n\n/**\n * Cross Entropy Loss.\n *\n * Computes the cross entropy loss between predictions and targets.\n * Commonly used for multi-class classification problems.\n *\n * Supports both integer class indices and one-hot encoded probabilities for targets.\n *\n * **Formula**: L = -mean(sum(target * log_softmax(input), dim=1))\n *\n * @param input - Predicted logits of shape (n_samples, n_classes)\n * @param target - True labels. Either:\n *                 - Class indices of shape (n_samples,)\n *                 - Probabilities/One-hot of shape (n_samples, n_classes)\n * @returns Scalar loss value (GradTensor)\n *\n * @example\n * ```ts\n * import { crossEntropyLoss } from 'deepbox/nn';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const pred = tensor([[0.7, 0.2, 0.1], [0.1, 0.8, 0.1]]);\n * const true_idx = tensor([0, 1]);\n * const loss = crossEntropyLoss(pred, true_idx);\n * ```\n */\nexport function crossEntropyLoss(input: Tensor, target: Tensor): number;\nexport function crossEntropyLoss(input: GradTensor, target: AnyTensor): GradTensor;\nexport function crossEntropyLoss(input: AnyTensor, target: AnyTensor): number | GradTensor {\n  const yPred = input instanceof GradTensor ? input : GradTensor.fromTensor(input);\n  const targetIsGrad = target instanceof GradTensor;\n  // Target usually doesn't require grad, but if it's soft labels (distillation), it might.\n  const yTrue =\n    target instanceof GradTensor ? target : GradTensor.fromTensor(target, { requiresGrad: false });\n\n  if (yPred.ndim !== 2) {\n    throw new ShapeError(`Input must be 2-dimensional (batch, classes); got ${yPred.ndim}`);\n  }\n\n  const nSamples = yPred.shape[0] ?? 0;\n  const nClasses = yPred.shape[1] ?? 0;\n\n  let targetTensor = yTrue;\n\n  // Handle class indices (1D)\n  if (yTrue.ndim === 1) {\n    if (targetIsGrad) {\n      throw new ShapeError(\"Target must be 2-dimensional when provided as GradTensor\");\n    }\n    if (yTrue.shape[0] !== nSamples) {\n      throw new ShapeError(\n        `Target must have same number of samples as input; got ${yTrue.shape[0]} and ${nSamples}`\n      );\n    }\n    // Convert to one-hot\n    // We need to access the underlying tensor data\n    const oneHot = toOneHot(yTrue.tensor, nClasses);\n    targetTensor = GradTensor.fromTensor(oneHot, { requiresGrad: false });\n  } else if (yTrue.ndim === 2) {\n    if (yTrue.shape[0] !== nSamples || yTrue.shape[1] !== nClasses) {\n      throw new ShapeError(\n        \"Target must be 1-dimensional class indices or have the same shape as input\"\n      );\n    }\n  } else {\n    throw new ShapeError(`Target must be 1D (indices) or 2D (probs); got ${yTrue.ndim}D`);\n  }\n\n  // Compute Log Softmax\n  const logProbs = logSoftmaxGrad(yPred, 1);\n\n  // Compute NLL: -sum(target * log_prob) / N\n  // Element-wise multiply\n  const weighted = logProbs.mul(targetTensor);\n\n  // Sum over classes (dim 1) -> (N,)\n  const sampleLoss = weighted.sum(1);\n\n  // Mean over batch -> scalar\n  // Note: sum(1) returns (N,), mean() returns scalar.\n  // But wait, `weighted` is negative log likelihood * target.\n  // For one-hot target, sum(target * log_prob) is log_prob[class].\n  // We want negative of that.\n  // And then mean.\n\n  const meanLoss = sampleLoss.mean().neg();\n  if (!(input instanceof GradTensor) && !targetIsGrad) {\n    const data = meanLoss.tensor.data;\n    if (Array.isArray(data)) {\n      throw new DTypeError(\"crossEntropyLoss does not support string dtype\");\n    }\n    if (data instanceof BigInt64Array) {\n      const raw = getBigIntElement(data, meanLoss.tensor.offset);\n      return Number(raw);\n    }\n    return getNumericElement(data, meanLoss.tensor.offset);\n  }\n  return meanLoss;\n}\n\n/**\n * Binary Cross Entropy Loss with logits.\n *\n * Combines sigmoid activation and binary cross entropy loss for numerical stability.\n *\n * @param input - Predicted logits of shape (n_samples,) or (n_samples, 1)\n * @param target - True binary labels of same shape as input\n * @returns Scalar loss value (GradTensor)\n */\nexport function binaryCrossEntropyWithLogitsLoss(input: Tensor, target: Tensor): number;\nexport function binaryCrossEntropyWithLogitsLoss(input: GradTensor, target: AnyTensor): GradTensor;\nexport function binaryCrossEntropyWithLogitsLoss(\n  input: AnyTensor,\n  target: AnyTensor\n): number | GradTensor {\n  const yPred = input instanceof GradTensor ? input : GradTensor.fromTensor(input);\n  const yTrue =\n    target instanceof GradTensor ? target : GradTensor.fromTensor(target, { requiresGrad: false });\n\n  // Check shapes\n  // Support (N,) and (N, 1)\n  let pred = yPred;\n  let truth = yTrue;\n\n  if (pred.ndim !== 1 && pred.ndim !== 2) {\n    throw new ShapeError(\"Input must be 1 or 2-dimensional\");\n  }\n  if (truth.ndim !== 1 && truth.ndim !== 2) {\n    throw new ShapeError(\"Target must be 1 or 2-dimensional\");\n  }\n\n  if (pred.ndim === 1) {\n    pred = pred.reshape([pred.shape[0] ?? 0, 1]);\n  }\n  if (truth.ndim === 1) {\n    truth = truth.reshape([truth.shape[0] ?? 0, 1]);\n  }\n\n  if (pred.ndim !== 2 || pred.shape[1] !== 1) {\n    throw new ShapeError(`Input must have shape (N,) or (N, 1)`);\n  }\n  if (truth.ndim !== 2 || truth.shape[1] !== 1) {\n    throw new ShapeError(`Target must be 1-dimensional or have shape (N, 1)`);\n  }\n  if ((pred.shape[0] ?? 0) !== (truth.shape[0] ?? 0)) {\n    throw new ShapeError(`Batch size mismatch`);\n  }\n\n  const predDtype = pred.dtype;\n  if (predDtype === \"string\") {\n    throw new DTypeError(\"Binary cross entropy does not support string dtype\");\n  }\n\n  // max(x, 0) - x * z + log(1 + exp(-abs(x)))\n  // We use autograd ops\n\n  // term1 = max(x, 0) -> relu(x)\n  const term1 = pred.relu();\n\n  // term2 = x * z\n  const term2 = pred.mul(truth);\n\n  // term3 = log(1 + exp(-abs(x)))\n  // abs(x) = relu(x) + relu(-x) or just use abs op if available?\n  // GradTensor doesn't have abs yet?\n  // We can use sign? Or just: abs(x) = sqrt(x^2)? No, gradient at 0.\n  // Or: abs(x) = max(x, -x).\n  // We have neg().\n  // We have max() (which I added).\n  // So abs(x) = max(x, x.neg())? No, max takes reduction axis.\n  // Elementwise max is not exposed on GradTensor yet.\n  // But wait, we can implement `softplus` (log(1+exp(x))) directly?\n  // The formula uses `log(1 + exp(-abs(x)))`.\n  // Alternative: `softplus(x)` if z=0, `softplus(x) - x` if z=1?\n  // Standard stable BCEWithLogits:\n  // loss = (1-z)*x + softplus(-x)   if x > 0\n  // loss = (1-z)*x + x + softplus(-x) ? No.\n\n  // PyTorch uses: max(x, 0) - x*z + log(1 + exp(-abs(x)))\n  // To implement `abs(x)` without primitive:\n  // `x.pow(2).sqrt()` is abs(x) but unstable grad at 0.\n  // `relu(x) + relu(-x)` works.\n\n  const negPred = pred.neg();\n  const absPred = pred.relu().add(negPred.relu());\n\n  const expNegAbs = absPred.neg().exp();\n  const scalarDtype = expNegAbs.dtype;\n  if (scalarDtype === \"string\") {\n    throw new DTypeError(\"binaryCrossEntropyWithLogitsLoss does not support string dtype\");\n  }\n  const one = GradTensor.scalar(1, { dtype: scalarDtype });\n  const term3 = one.add(expNegAbs).log();\n\n  const loss = term1.sub(term2).add(term3).mean();\n  if (!(input instanceof GradTensor) && !(target instanceof GradTensor)) {\n    const data = loss.tensor.data;\n    if (Array.isArray(data)) {\n      throw new DTypeError(\"binaryCrossEntropyWithLogitsLoss does not support string dtype\");\n    }\n    if (data instanceof BigInt64Array) {\n      const raw = getBigIntElement(data, loss.tensor.offset);\n      return Number(raw);\n    }\n    return getNumericElement(data, loss.tensor.offset);\n  }\n  return loss;\n}\n","import { DTypeError, getElementAsNumber, InvalidParameterError, ShapeError } from \"../../core\";\nimport {\n  abs,\n  add,\n  clip,\n  log,\n  mean,\n  mul,\n  neg,\n  pow,\n  sqrt,\n  sub,\n  sum,\n  Tensor,\n  tensor,\n} from \"../../ndarray\";\nimport { offsetFromFlatIndex } from \"../../ndarray/tensor/strides\";\nimport { computeStrides } from \"../../ndarray/tensor/Tensor\";\n\nexport {\n  binaryCrossEntropyWithLogitsLoss,\n  crossEntropyLoss,\n} from \"./crossEntropy\";\n\nfunction shapesEqual(a: readonly number[], b: readonly number[]): boolean {\n  if (a.length !== b.length) return false;\n  for (let i = 0; i < a.length; i++) {\n    if ((a[i] ?? 0) !== (b[i] ?? 0)) return false;\n  }\n  return true;\n}\n\nfunction ensureSameShape(a: Tensor, b: Tensor, context: string): void {\n  if (!shapesEqual(a.shape, b.shape)) {\n    throw new ShapeError(`Shape mismatch in ${context}: [${a.shape}] vs [${b.shape}]`);\n  }\n}\n\nfunction ensureNumeric(t: Tensor, context: string): void {\n  if (t.dtype === \"string\") {\n    throw new DTypeError(`${context} does not support string dtype`);\n  }\n}\n\ntype NumericTensorData = Exclude<Tensor[\"data\"], string[]>;\n\nfunction validateReduction(reduction: \"mean\" | \"sum\" | \"none\", context: string): void {\n  if (reduction !== \"mean\" && reduction !== \"sum\" && reduction !== \"none\") {\n    throw new InvalidParameterError(\n      `${context} reduction must be 'mean', 'sum', or 'none'`,\n      \"reduction\",\n      reduction\n    );\n  }\n}\n\nfunction readNumericFlat(\n  data: NumericTensorData,\n  flat: number,\n  logicalStrides: readonly number[],\n  strides: readonly number[],\n  offset: number\n): number {\n  const dataOffset = offsetFromFlatIndex(flat, logicalStrides, strides, offset);\n  return getElementAsNumber(data, dataOffset);\n}\n\n/**\n * Mean Squared Error (MSE) loss function.\n *\n * **Mathematical Formula:**\n * ```\n * MSE = mean((y_pred - y_true)^2)\n * ```\n *\n * **Use Cases:**\n * - Regression tasks\n * - Continuous value prediction\n * - Measuring distance between predictions and targets\n *\n * **Properties:**\n * - Always non-negative\n * - Penalizes large errors more heavily (quadratic)\n * - Differentiable everywhere\n *\n * @param predictions - Predicted values\n * @param targets - True target values\n * @param reduction - How to reduce the loss: 'mean', 'sum', or 'none'\n * @returns Scalar loss value (or tensor if reduction='none')\n *\n * @example\n * ```ts\n * import { mseLoss } from 'deepbox/nn/losses';\n * import { tensor } from 'deepbox/ndarray';\n *\n * const predictions = tensor([2.5, 0.0, 2.1, 7.8]);\n * const targets = tensor([3.0, -0.5, 2.0, 8.0]);\n * const loss = mseLoss(predictions, targets); // Scalar tensor\n * ```\n *\n * @category Loss Functions\n */\nexport function mseLoss(\n  predictions: Tensor,\n  targets: Tensor,\n  reduction: \"mean\" | \"sum\" | \"none\" = \"mean\"\n): Tensor {\n  validateReduction(reduction, \"mseLoss\");\n  ensureNumeric(predictions, \"mseLoss\");\n  ensureNumeric(targets, \"mseLoss\");\n  ensureSameShape(predictions, targets, \"mseLoss\");\n\n  const diff = sub(predictions, targets);\n  const squaredDiff = pow(diff, tensor(2, { dtype: diff.dtype, device: diff.device }));\n\n  if (reduction === \"none\") {\n    return squaredDiff;\n  }\n  if (reduction === \"sum\") {\n    return sum(squaredDiff);\n  }\n  return mean(squaredDiff);\n}\n\n/**\n * Mean Absolute Error (MAE) loss function, also known as L1 loss.\n *\n * **Mathematical Formula:**\n * ```\n * MAE = mean(|y_pred - y_true|)\n * ```\n *\n * **Use Cases:**\n * - Regression tasks where outliers should have less influence\n * - More robust to outliers than MSE\n *\n * **Properties:**\n * - Always non-negative\n * - Linear penalty for errors\n * - Less sensitive to outliers than MSE\n *\n * @param predictions - Predicted values\n * @param targets - True target values\n * @param reduction - How to reduce the loss: 'mean', 'sum', or 'none'\n * @returns Scalar loss value (or tensor if reduction='none')\n *\n * @category Loss Functions\n */\nexport function maeLoss(\n  predictions: Tensor,\n  targets: Tensor,\n  reduction: \"mean\" | \"sum\" | \"none\" = \"mean\"\n): Tensor {\n  validateReduction(reduction, \"maeLoss\");\n  ensureNumeric(predictions, \"maeLoss\");\n  ensureNumeric(targets, \"maeLoss\");\n  ensureSameShape(predictions, targets, \"maeLoss\");\n\n  const diff = sub(predictions, targets);\n  const absDiff = abs(diff);\n\n  if (reduction === \"none\") {\n    return absDiff;\n  }\n  if (reduction === \"sum\") {\n    return sum(absDiff);\n  }\n  return mean(absDiff);\n}\n\n/**\n * Binary Cross-Entropy (BCE) loss function.\n *\n * **Mathematical Formula:**\n * ```\n * BCE = -mean(y_true * log(y_pred) + (1 - y_true) * log(1 - y_pred))\n * ```\n *\n * **Use Cases:**\n * - Binary classification tasks\n * - Multi-label classification (independent binary decisions)\n * - Predictions should be probabilities in (0, 1)\n *\n * **Properties:**\n * - Requires predictions in range (0, 1) - use sigmoid activation\n * - Targets should be 0 or 1\n * - Numerically stable with epsilon for log\n *\n * @param predictions - Predicted probabilities (0 to 1)\n * @param targets - True binary labels (0 or 1)\n * @param reduction - How to reduce the loss: 'mean', 'sum', or 'none'\n * @returns Scalar loss value (or tensor if reduction='none')\n *\n * @category Loss Functions\n */\nexport function binaryCrossEntropyLoss(\n  predictions: Tensor,\n  targets: Tensor,\n  reduction: \"mean\" | \"sum\" | \"none\" = \"mean\"\n): Tensor {\n  validateReduction(reduction, \"binaryCrossEntropyLoss\");\n  ensureNumeric(predictions, \"binaryCrossEntropyLoss\");\n  ensureNumeric(targets, \"binaryCrossEntropyLoss\");\n  ensureSameShape(predictions, targets, \"binaryCrossEntropyLoss\");\n\n  const epsilon = 1e-7;\n  const predClamped = clip(predictions, epsilon, 1 - epsilon);\n\n  const logPred = log(predClamped);\n  const term1 = mul(targets, logPred);\n\n  const one = tensor(1, {\n    dtype: predictions.dtype === \"float64\" ? \"float64\" : \"float32\",\n    device: predictions.device,\n  });\n  const oneMinusTargets = sub(one, targets);\n  const oneMinusPred = sub(one, predClamped);\n  const logOneMinusPred = log(oneMinusPred);\n  const term2 = mul(oneMinusTargets, logOneMinusPred);\n\n  const loss = neg(add(term1, term2));\n\n  if (reduction === \"none\") {\n    return loss;\n  }\n  if (reduction === \"sum\") {\n    return sum(loss);\n  }\n  return mean(loss);\n}\n\n/**\n * Root Mean Squared Error (RMSE) loss function.\n *\n * **Mathematical Formula:**\n * ```\n * RMSE = sqrt(mean((y_pred - y_true)^2))\n * ```\n *\n * **Use Cases:**\n * - Regression tasks\n * - When you want error in same units as target\n * - More interpretable than MSE\n *\n * @param predictions - Predicted values\n * @param targets - True target values\n * @returns Scalar loss value\n *\n * @category Loss Functions\n */\nexport function rmseLoss(predictions: Tensor, targets: Tensor): Tensor {\n  ensureNumeric(predictions, \"rmseLoss\");\n  ensureNumeric(targets, \"rmseLoss\");\n  ensureSameShape(predictions, targets, \"rmseLoss\");\n\n  const mse = mseLoss(predictions, targets, \"mean\");\n  return sqrt(mse);\n}\n\n/**\n * Huber loss function - combines MSE and MAE.\n *\n * **Mathematical Formula:**\n * ```\n * Huber(a) = 0.5 * a^2           if |a| <= delta\n *          = delta * (|a| - 0.5 * delta)  otherwise\n * where a = y_pred - y_true\n * ```\n *\n * **Use Cases:**\n * - Regression with outliers\n * - Robust to outliers while maintaining MSE benefits for small errors\n *\n * **Properties:**\n * - Quadratic for small errors (like MSE)\n * - Linear for large errors (like MAE)\n * - Controlled by delta parameter\n *\n * @param predictions - Predicted values\n * @param targets - True target values\n * @param delta - Threshold where loss transitions from quadratic to linear\n * @param reduction - How to reduce the loss: 'mean', 'sum', or 'none'\n * @returns Scalar loss value (or tensor if reduction='none')\n *\n * @category Loss Functions\n */\nexport function huberLoss(\n  predictions: Tensor,\n  targets: Tensor,\n  delta = 1.0,\n  reduction: \"mean\" | \"sum\" | \"none\" = \"mean\"\n): Tensor {\n  validateReduction(reduction, \"huberLoss\");\n  ensureNumeric(predictions, \"huberLoss\");\n  ensureNumeric(targets, \"huberLoss\");\n  ensureSameShape(predictions, targets, \"huberLoss\");\n\n  if (!Number.isFinite(delta) || delta <= 0) {\n    throw new InvalidParameterError(`delta must be positive; got ${delta}`, \"delta\", delta);\n  }\n\n  const diff = sub(predictions, targets);\n  const absDiff = abs(diff);\n\n  const absData = absDiff.data;\n  if (Array.isArray(absData)) {\n    throw new DTypeError(\"huberLoss does not support string dtype\");\n  }\n  const dtype = predictions.dtype === \"float64\" ? \"float64\" : \"float32\";\n  const lossData = dtype === \"float64\" ? new Float64Array(diff.size) : new Float32Array(diff.size);\n  const logicalStrides = computeStrides(absDiff.shape);\n  for (let i = 0; i < diff.size; i++) {\n    const absVal = readNumericFlat(absData, i, logicalStrides, absDiff.strides, absDiff.offset);\n    if (absVal <= delta) {\n      lossData[i] = 0.5 * absVal * absVal;\n    } else {\n      lossData[i] = delta * (absVal - 0.5 * delta);\n    }\n  }\n\n  const loss = Tensor.fromTypedArray({\n    data: lossData,\n    shape: predictions.shape,\n    dtype,\n    device: predictions.device,\n  });\n\n  if (reduction === \"none\") {\n    return loss;\n  }\n  if (reduction === \"sum\") {\n    return sum(loss);\n  }\n  return mean(loss);\n}\n"]}