@genai-fi/nanogpt 0.6.3 → 0.7.0

package/dist/{random_width-DWzaOgrn.js → random_width-DkYP8W8N.js}

@@ -1,25 +1,24 @@
-import { o as T, q as I, E as O, bB as Oa, bC as Ma, bD as Ci, i as b, bE as Ii, L, bF as Di, bG as $i, bH as Ti, bI as zi, f as Ei, bJ as Li, bK as Fi, bL as Oi, bM as Mi, bN as _a, bO as _i, bP as Ra, bQ as Ri, bR as Ba, bS as Bi, B as Ge, h as kt, bn as Wa, bT as Wi, bU as Gi, bV as Pi, l as Pe, b as V, a as w, bW as Ga, bX as Ui, bY as Vi, bZ as ji, p as ce, a5 as pt, bs as Pa, b_ as Ki, b$ as Hi, c0 as qi, c1 as Ji, c2 as Zi, bu as Yi, c3 as Xi, c4 as Qi, x as Ua, ah as Va, ao as ja, c5 as tr, c6 as Ka, a7 as z, c7 as Ps, c8 as Ha, c9 as qa, g as pn, ca as Us, cb as Ja, cc as Za, cd as Ya, A as Xa, ce as Qa, cf as tl, cg as el, bc as nl, ch as sl, a4 as he, s as et, aj as U, ci as il, bj as rl, au as ht, cj as ol, z as Q, ck as al, cl as ll, cm as ul, cn as cl, co as hl, cp as pl, cq as dl, cr as fl, C as ml, cs as gl, bg as bl, bm as yl, ct as wl, J as kl, cu as xl, a0 as Nl, cv as vl, cw as Al, cx as Sl, aq as Cl, cy as Il, a8 as Dl, aS as $l, bo as Tl, ag as zl, bp as El, G as Ll, aU as Fl, ak as Ol, cz as Ml, cA as _l, cB as Rl, al as Bl, a9 as Wl, cC as Gl, cD as Pl, cE as Ul, M as Vl, bq as jl, cF as Kl, cG as Hl, a2 as ql, a_ as Jl, cH as Zl, cI as Yl, br as Xl, aY as Ql, P as tu, cJ as eu, j as rs, an as nu, bt as su, ax as iu, K as ru, as as ou, ar as au, D as lu, b9 as uu, cK as cu, ba as hu, cL as pu, b0 as du, aQ as fu, ap as mu, cM as gu, a1 as bu, aO as yu, S as wu, F as ku, bw as xu, cN as Nu, bx as vu, at as Au, bz as Su, N as Cu, cO as Iu, I as Du, b2 as $u, b1 as Tu, cP as Oe, cQ as zu, c as Eu, af as Vs, cR as Lu, t as x, a3 as $e, cS as S, cT as He, cU as qe, ai as Vt, d as J, ac as Fu, cV as js, k as Jt, H as Ou, T as Te, cW as Mu, cX as _u, m as Ks, cY as Ru, cZ as Hs, c_ as Bu } from "./index-BAzbokzv.js";
-import { k as ke, c as Nt, o as ze, s as er, b as nr, d as Wu, m as sr, t as In, l as ir, v as os, a as Gu, S as Pu, p as Uu, w as as, x as rr, y as Je, z as Vu, A as ju } from "./selu_util-sntGesxr.js";
-import { s as Dt, n as wt, w as ne, a as Ze, g as Ye, b as ls, c as K, d as Ce, e as Yt, f as Ku, u as dn, h as ye, t as fn, i as Hu, l as qu, j as us, m as or, k as qt } from "./ops-C0sQEcPw.js";
-import { r as N } from "./reshape-krWGKraP.js";
-import { s as W } from "./sum-BE_Irnim.js";
-import { m as ct } from "./mat_mul-7121rsJk.js";
-import { s as Qt } from "./split-eiktj-6L.js";
-import { s as Ju, c as ar } from "./sin-D_h-qCSx.js";
-import { e as Hn, g as lr, h as cs, c as Zu } from "./axis_util-Bu4h7XWV.js";
-import { a as se, e as ie, l as Yu } from "./log_sum_exp-YEo2h3gb.js";
-import { s as Dn } from "./stack-dfEEz2OY.js";
-import { o as xe } from "./ones-BBlSRqn1.js";
-import { M as Xu, f as ur, r as Qu, d as tc, a as $n } from "./dropout-Bciw46HT.js";
-import { z as vt } from "./zeros-YJDE7oRb.js";
-import { c as pe } from "./concat-5aPGqw3Z.js";
-import { g as cr } from "./gather-DjyCjmOD.js";
-import { s as hr } from "./softmax-fsdtf6JC.js";
-import { m as Ee } from "./max-DtlIuVeW.js";
-import { t as ec } from "./tensor-Xyi595sG.js";
-import { r as nc } from "./range-DYsrnfiy.js";
-import { m as sc } from "./norm-CzltS9Fz.js";
-import { v as ic } from "./variable-wSS22xj5.js";
+import { x as T, y as I, E as O, bE as Oa, bF as Ma, bG as Ci, l as b, bH as Ii, Q as L, bI as Di, bJ as $i, bK as Ti, bL as zi, h as Ei, bM as Li, bN as Fi, bO as Oi, bP as Mi, bQ as _a, bR as _i, bS as Ra, bT as Ri, bU as Ba, bV as Bi, F as Ge, k as kt, bq as Wa, bW as Wi, bX as Gi, bY as Pi, q as Pe, c as V, a as w, bZ as Ga, b_ as Ui, b$ as Vi, c0 as ji, p as ce, aU as pt, bv as Pa, c1 as Ki, c2 as Hi, c3 as qi, c4 as Ji, c5 as Zi, bx as Xi, c6 as Yi, c7 as Qi, C as Ua, ai as Va, ao as ja, c8 as tr, c9 as Ka, a6 as z, ca as Ps, cb as Ha, cc as qa, j as pn, cd as Us, ce as Ja, cf as Za, cg as Xa, A as Ya, ch as Qa, ci as tl, cj as el, bf as nl, ck as sl, aT as he, b as et, aj as U, cl as il, bm as rl, au as ht, cm as ol, z as Q, cn as al, co as ll, cp as ul, cq as cl, cr as hl, cs as pl, ct as dl, cu as fl, H as ml, cv as gl, bj as bl, bp as yl, cw as wl, M as kl, cx as xl, a3 as Nl, cy as vl, cz as Al, cA as Sl, aq as Cl, cB as Il, a7 as Dl, aV as $l, br as Tl, ah as zl, bs as El, G as Ll, aX as Fl, ak as Ol, cC as Ml, cD as _l, cE as Rl, al as Bl, a8 as Wl, cF as Gl, cG as Pl, cH as Ul, N as Vl, bt as jl, cI as Kl, cJ as Hl, aR as ql, b1 as Jl, cK as Zl, cL as Xl, bu as Yl, a$ as Ql, P as tu, cM as eu, n as rs, an as nu, bw as su, ax as iu, O as ru, as as ou, ar as au, I as lu, bc as uu, cN as cu, bd as hu, cO as pu, b3 as du, aP as fu, ap as mu, cP as gu, a4 as bu, aN as yu, S as wu, J as ku, bz as xu, cQ as Nu, bA as vu, at as Au, bC as Su, U as Cu, cR as Iu, L as Du, b5 as $u, b4 as Tu, cS as Oe, cT as zu, i as Eu, ag as Vs, cU as Lu, t as x, aS as $e, cV as S, cW as He, cX as qe, ab as Vt, d as J, ac as Fu, cY as js, o as Jt, K as Ou, T as Te, cZ as Mu, c_ as _u, m as Ks, c$ as Ru, d0 as Hs, d1 as Bu } from "./index-C0dhsYom.js";
+import { k as ke, c as Nt, o as ze, s as er, b as nr, d as Wu, m as sr, t as In, l as ir, v as os, a as Gu, S as Pu, p as Uu, w as as, x as rr, y as Je, z as Vu, A as ju } from "./selu_util-Dt_iuXaq.js";
+import { s as Dt, n as wt, w as ne, a as Ze, g as Xe, b as ls, t as K, c as Ce, d as Xt, e as Ku, u as dn, f as ye, h as fn, i as Hu, l as qu, j as us, m as or, k as qt, o as Ju } from "./ops-CIQLNshk.js";
+import { r as N } from "./reshape-BMUzc1UY.js";
+import { s as W } from "./sum-B8wEpKsg.js";
+import { m as ct } from "./mat_mul-1nwdPkQ_.js";
+import { s as Qt } from "./split-CHzJjxDv.js";
+import { s as Zu, c as ar } from "./sin-Cocju-BY.js";
+import { e as Hn, g as lr, h as cs, c as Xu } from "./axis_util-DaAl5MER.js";
+import { a as se, e as ie, l as Yu } from "./log_sum_exp-D086OgZJ.js";
+import { s as Dn } from "./stack-Dpgg_1W1.js";
+import { o as xe } from "./ones-D63HpSF_.js";
+import { M as Qu, f as ur, r as tc, d as ec, a as $n } from "./dropout-hpDwECTe.js";
+import { z as vt } from "./zeros-DCPCdFGq.js";
+import { c as pe } from "./concat-CsHeR4zV.js";
+import { g as cr } from "./gather-D0_gPiBz.js";
+import { s as hr } from "./softmax-GPNK3o-U.js";
+import { m as Ee } from "./max-BQc2Aj-I.js";
+import { t as nc } from "./tensor-RvZVNmg0.js";
+import { r as sc } from "./range-CYzpQY53.js";
+import { v as ic } from "./variable-DXEUOwew.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -531,13 +530,13 @@ const dr = /* @__PURE__ */ T({ depthwiseConv2d_: Zc });
  * limitations under the License.
  * =============================================================================
  */
-function Yc(n, t) {
+function Xc(n, t) {
   let e = I(n, "a", "equal", "string_or_numeric"), s = I(t, "b", "equal", "string_or_numeric");
   [e, s] = Ge(e, s), kt(e.shape, s.shape);
   const i = { a: e, b: s };
   return O.runKernel(Wa, i);
 }
-const re = /* @__PURE__ */ T({ equal_: Yc });
+const re = /* @__PURE__ */ T({ equal_: Xc });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -554,13 +553,13 @@ const re = /* @__PURE__ */ T({ equal_: Yc });
  * limitations under the License.
  * =============================================================================
  */
-function Xc(n) {
+function Yc(n) {
   let t = I(n, "x", "erf");
   b(t.dtype === "int32" || t.dtype === "float32", () => "Input dtype must be `int32` or `float32`."), t.dtype === "int32" && (t = L(t, "float32"));
   const e = { x: t };
   return O.runKernel(Wi, e);
 }
-const Qc = /* @__PURE__ */ T({ erf_: Xc });
+const Qc = /* @__PURE__ */ T({ erf_: Yc });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -900,7 +899,7 @@ const gn = /* @__PURE__ */ T({ reverse_: kh });
  */
 function xh(n) {
   const e = { x: I(n, "x", "rsqrt", "float32") };
-  return O.runKernel(Yi, e);
+  return O.runKernel(Xi, e);
 }
 const Nh = /* @__PURE__ */ T({ rsqrt_: xh });
 /**
@@ -921,7 +920,7 @@ const Nh = /* @__PURE__ */ T({ rsqrt_: xh });
  */
 function vh(n) {
   const e = { x: I(n, "x", "selu") };
-  return O.runKernel(Xi, e);
+  return O.runKernel(Yi, e);
 }
 const Ah = /* @__PURE__ */ T({ selu_: vh });
 function Sh(n, t, e, s, i, r = [1, 1], o = "NHWC") {
@@ -1060,7 +1059,7 @@ const bn = /* @__PURE__ */ T({ slice4d_: Eh });
 function Lh(n, t = 0, e = 1, s, i) {
   if (Ua(n), s != null && s === "bool")
     throw new Error("Unsupported data type $ { dtype }");
-  const r = new Xu(t, e, s, !0, i), o = Va(n, s);
+  const r = new Qu(t, e, s, !0, i), o = Va(n, s);
   for (let a = 0; a < o.values.length; a++)
     o.values[a] = r.nextValue();
   return o.toTensor();
@@ -1305,7 +1304,7 @@ function Uh({ a: n, b: t, transposeA: e = !1, transposeB: s = !1, bias: i, activ
     return at([ut, ft, bt, mt]), { value: N(bt, C), gradFunc: E };
   })(v, D, $);
 }
-const Ys = /* @__PURE__ */ T({ fusedMatMul_: Uh });
+const Xs = /* @__PURE__ */ T({ fusedMatMul_: Uh });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -1402,7 +1401,7 @@ class Vh {
    * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}
    */
   static rmsprop(t, e = 0.9, s = 0, i = null, r = !1) {
-    return new Ya(t, e, s, i, r);
+    return new Xa(t, e, s, i, r);
   }
   /**
    * Constructs a `tf.AdamOptimizer` that uses the Adam algorithm.
@@ -1417,7 +1416,7 @@ class Vh {
    * @doc {heading: 'Training', subheading: 'Optimizers', namespace: 'train'}
    */
   static adam(t = 1e-3, e = 0.9, s = 0.999, i = null) {
-    return new Xa(t, e, s, i);
+    return new Ya(t, e, s, i);
   }
   /**
    * Constructs a `tf.AdadeltaOptimizer` that uses the Adadelta algorithm.
@@ -1663,7 +1662,7 @@ const Zh = {
  * limitations under the License.
  * =============================================================================
  */
-const Yh = {
+const Xh = {
   kernelName: Ci,
   inputsToSave: ["x"],
   gradFunc: (n, t) => {
@@ -1687,7 +1686,7 @@ const Yh = {
  * limitations under the License.
  * =============================================================================
  */
-const Xh = {
+const Yh = {
   kernelName: al,
   inputsToSave: ["x"],
   gradFunc: (n, t) => {
@@ -2092,7 +2091,7 @@ const mp = {
   gradFunc: (n, t, e) => {
     const [s] = t, { clipValueMin: i, clipValueMax: r } = e;
     return {
-      x: () => ne(Ze(Ye(s, i), ls(s, r)), n, Q(n))
+      x: () => ne(Ze(Xe(s, i), ls(s, r)), n, Q(n))
     };
   }
 };
@@ -2270,7 +2269,7 @@ const vp = {
   inputsToSave: ["x"],
   gradFunc: (n, t) => {
     const [e] = t;
-    return { x: () => w(wt(Ju(L(e, "float32"))), n) };
+    return { x: () => w(wt(Zu(L(e, "float32"))), n) };
   }
 };
 /**
@@ -2626,7 +2625,7 @@ const Mp = {
   inputsToSave: ["x", "indices"],
   gradFunc: (n, t, e) => {
     const [s, i] = t, { axis: r, batchDims: o } = e, a = ce(r, s.shape)[0], l = (u, c, h) => () => {
-      const p = u.shape, f = c.size, y = p.slice(0, a), g = y.length, m = p.slice(r, p.length).slice(1), A = m.length, k = Xs(0, g), C = Xs(g + 1, g + 1 + A), v = Qs([
+      const p = u.shape, f = c.size, y = p.slice(0, a), g = y.length, m = p.slice(r, p.length).slice(1), A = m.length, k = Ys(0, g), C = Ys(g + 1, g + 1 + A), v = Qs([
         y,
         [f],
         m
@@ -2642,7 +2641,7 @@ const Mp = {
       return { x: l(s, i, n), indices: () => i };
   }
 };
-function Xs(n, t) {
+function Ys(n, t) {
   const e = [];
   for (let s = n; s < t; ++s)
     e.push(s);
@@ -2779,7 +2778,7 @@ const Pp = {
   kernelName: Bl,
   inputsToSave: ["x"],
   gradFunc: (n, t, e) => {
-    const [s] = t, { alpha: i } = e, r = Yt(s, 0);
+    const [s] = t, { alpha: i } = e, r = Xt(s, 0);
     return { x: () => ne(r, n, w(n, i)) };
   }
 };
@@ -2978,7 +2977,7 @@ const Jp = {
   inputsToSave: ["a", "b"],
   gradFunc: (n, t) => {
     const [e, s] = t;
-    return { a: () => w(n, L(Ye(e, s), "float32")), b: () => w(n, L(Ku(e, s), "float32")) };
+    return { a: () => w(n, L(Xe(e, s), "float32")), b: () => w(n, L(Ku(e, s), "float32")) };
   }
 };
 /**
@@ -3016,7 +3015,7 @@ function Zp(n, t, e, s, i, r, o) {
   const y = { dy: c, input: h, output: p }, g = { filterSize: s, strides: i, pad: r, dimRoundingMode: o }, m = O.runKernel(Kl, y, g);
   return f ? N(m, [m.shape[1], m.shape[2], m.shape[3], m.shape[4]]) : m;
 }
-const Yp = /* @__PURE__ */ T({ maxPool3dGrad_: Zp });
+const Xp = /* @__PURE__ */ T({ maxPool3dGrad_: Zp });
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -3033,14 +3032,14 @@ const Yp = /* @__PURE__ */ T({ maxPool3dGrad_: Zp });
  * limitations under the License.
  * =============================================================================
  */
-const Xp = {
+const Yp = {
   kernelName: Vi,
   inputsToSave: ["x"],
   outputsToSave: [!0],
   gradFunc: (n, t, e) => {
     const [s, i] = t, { filterSize: r, strides: o, pad: a, dimRoundingMode: l } = e;
     return {
-      x: () => Yp(n, s, i, r, o, a, l)
+      x: () => Xp(n, s, i, r, o, a, l)
     };
   }
 };
@@ -3114,7 +3113,7 @@ const nd = {
   kernelName: ji,
   inputsToSave: ["x"],
   gradFunc: (n, t, e) => {
-    const [s] = t, { axis: i } = e, r = ce(i, s.shape), a = Zu(s.shape, r)[1], l = pn(a);
+    const [s] = t, { axis: i } = e, r = ce(i, s.shape), a = Xu(s.shape, r)[1], l = pn(a);
     return { x: () => {
       const c = s.shape.slice();
       r.forEach((f) => {
@@ -3173,7 +3172,7 @@ const id = {
   inputsToSave: ["a", "b"],
   gradFunc: (n, t) => {
     const [e, s] = t;
-    return { a: () => w(n, L(ls(e, s), "float32")), b: () => w(n, L(Yt(e, s), "float32")) };
+    return { a: () => w(n, L(ls(e, s), "float32")), b: () => w(n, L(Xt(e, s), "float32")) };
   }
 };
 /**
@@ -3217,7 +3216,7 @@ const rd = {
  * =============================================================================
  */
 const od = {
-  kernelName: Yl,
+  kernelName: Xl,
   inputsToSave: ["a", "b"],
   gradFunc: (n, t) => {
     const [e, s] = t, i = kt(e.shape, s.shape);
@@ -3247,7 +3246,7 @@ const od = {
  * =============================================================================
  */
 const ad = {
-  kernelName: Xl,
+  kernelName: Yl,
   inputsToSave: ["a", "b"],
   gradFunc: (n, t) => {
     const [e, s] = t, i = kt(e.shape, s.shape);
@@ -3400,7 +3399,7 @@ const pd = {
       const p = ht(r.shape, a);
       return p.length > 0 && (h = W(h, p)), N(h, r.shape);
     }, b: () => {
-      const c = Yt(r, 0), h = ne(c, se(r), Q(r));
+      const c = Xt(r, 0), h = ne(c, se(r), Q(r));
       let p = w(n, w(i, h));
       const f = ht(o.shape, a);
       return f.length > 0 && (p = W(p, f)), N(p, o.shape);
@@ -3427,7 +3426,7 @@ const dd = {
   kernelName: nu,
   inputsToSave: ["x", "alpha"],
   gradFunc: (n, t) => {
-    const [e, s] = t, i = Yt(e, 0);
+    const [e, s] = t, i = Xt(e, 0);
     return {
       x: () => ne(i, n, w(n, s)),
       alpha: () => {
@@ -3726,7 +3725,7 @@ const Sd = {
  * =============================================================================
  */
 const Cd = {
-  kernelName: Yi,
+  kernelName: Xi,
   inputsToSave: ["x"],
   gradFunc: (n, t) => {
     const [e] = t;
@@ -3780,13 +3779,13 @@ const Id = {
  * =============================================================================
  */
 const Dd = {
-  kernelName: Xi,
+  kernelName: Yi,
   inputsToSave: ["x"],
   gradFunc: (n, t) => {
     const [e] = t;
     return {
       x: () => {
-        const s = Yt(e, et(0)), i = et(Gu), r = et(Pu), o = w(n, r), a = w(w(n, i), ie(L(e, "float32")));
+        const s = Xt(e, et(0)), i = et(Gu), r = et(Pu), o = w(n, r), a = w(w(n, i), ie(L(e, "float32")));
         return ne(s, o, a);
       }
     };
@@ -4333,7 +4332,7 @@ const Hd = {
 };
 function qd(n, t) {
   const e = Oe(t, Q(t)), s = cr(n, e);
-  let i = Ye(t, et(0, "int32"));
+  let i = Xe(t, et(0, "int32"));
   const r = s.rank - i.rank;
   for (let a = 0; a < r; ++a)
     i = ye(i, a + 1);
@@ -4383,8 +4382,8 @@ const Zd = [
   qh,
   Jh,
   Zh,
-  Yh,
   Xh,
+  Yh,
   Qh,
   tp,
   ep,
@@ -4431,7 +4430,7 @@ const Zd = [
   ti,
   ti,
   Jp,
-  Xp,
+  Yp,
   ed,
   nd,
   sd,
@@ -4636,7 +4635,7 @@ function Jn(n) {
       }
     }
 }
-function Xe(n, t = {}, e = {}, s = "object", i = !1) {
+function Ye(n, t = {}, e = {}, s = "object", i = !1) {
   if (typeof n == "string") {
     const r = n;
     let o;
@@ -4683,11 +4682,11 @@ function Xe(n, t = {}, e = {}, s = "object", i = !1) {
     }
   }
 }
-function Yd(n, t) {
+function Xd(n, t) {
   return n < t ? -1 : n > t ? 1 : 0;
 }
 function ln(n, t) {
-  return -1 * Yd(n, t);
+  return -1 * Xd(n, t);
 }
 function te(n) {
   if (n == null)
@@ -4697,7 +4696,7 @@ function te(n) {
     t.indexOf(e) === -1 && t.push(e);
   return t;
 }
-function Xd(n) {
+function Yd(n) {
   if (n == null)
     throw new d(`Invalid value in obj: ${JSON.stringify(n)}`);
   for (const t in n)
@@ -5023,7 +5022,7 @@ function Zn(n, t) {
   return Ce(n, t);
 }
 function zn(n, t = 0, e = 1, s, i) {
-  return Qu(n, t, e, s, i);
+  return tc(n, t, e, s, i);
 }
 function Ut(n, t, e, s) {
   if (n.rank < 2 || t.rank < 2)
@@ -5034,12 +5033,12 @@ function Ut(n, t, e, s) {
       throw new B(`If rank y >= 3, then the second last dim of y must equal the last dim of x but got x shape = ${n.shape} and  y shape = ${t.shape}`);
   }
   if (n.rank === 2 && t.rank === 2)
-    return Ys({
+    return Xs({
       a: n,
       b: t,
       transposeA: !1,
       transposeB: !1,
-      bias: s ? Yn(n.rank, s, _t()) : null,
+      bias: s ? Xn(n.rank, s, _t()) : null,
       activation: e
     });
   {
@@ -5048,12 +5047,12 @@ function Ut(n, t, e, s) {
     const o = t.shape.slice(), a = o.pop(), l = o.pop(), u = [...o, a], c = Array.from({ length: t.rank }, (y, g) => g === 0 ? t.rank - 2 : g <= t.rank - 2 ? g - 1 : g);
     t = N(K(t, c), [l, -1]);
     const h = [...i, ...u];
-    return N(Ys({
+    return N(Xs({
       a: n,
       b: t,
       transposeA: !1,
       transposeB: !1,
-      bias: s ? Yn(n.rank, s, _t()) : null,
+      bias: s ? Xn(n.rank, s, _t()) : null,
       activation: e
     }), h);
   }
@@ -5064,7 +5063,7 @@ function Ir(n, t, e) {
 function tn(n) {
   return w(n, n);
 }
-function Yn(n, t, e) {
+function Xn(n, t, e) {
   const s = t.shape;
   if (t.rank !== 1 && t.rank !== n)
     throw new d(`Unexpected bias dimensions: ${t.rank}; expected it to be 1 or ${n}`);
@@ -5088,7 +5087,7 @@ function Yn(n, t, e) {
   throw new d(`Unsupported input rank by biasAdd: ${t.rank}`);
 }
 function Rt(n, t, e) {
-  return x(() => (e == null && (e = _t()), tt(e), z(n, Yn(n.rank, t, e))));
+  return x(() => (e == null && (e = _t()), tt(e), z(n, Xn(n.rank, t, e))));
 }
 function ff(n, t = 1) {
   if (t !== 1)
@@ -5099,7 +5098,7 @@ function mf(n) {
   return x(() => U(n, z($e(n), 1)));
 }
 function Dr(n, t, e, s) {
-  return x(() => tc(n, t, e, s));
+  return x(() => ec(n, t, e, s));
 }
 function gf(n) {
   return x(() => {
@@ -5438,9 +5437,9 @@ const ai = {
   zeros: "Zeros"
 };
 function li(n, t = {}) {
-  return Xe(n, qe.getMap().classNameMap, t, "initializer");
+  return Ye(n, qe.getMap().classNameMap, t, "initializer");
 }
-function X(n) {
+function Y(n) {
   return ks(n);
 }
 function Z(n) {
@@ -5473,7 +5472,7 @@ function Z(n) {
  * https://opensource.org/licenses/MIT.
  * =============================================================================
  */
-function Xn(n) {
+function Yn(n) {
   return Array.isArray(n) && Array.isArray(n[0]);
 }
 function yn(n) {
@@ -6438,14 +6437,14 @@ function zf(n, t) {
     throw new d(`The dtype of the feed (${t.dtype}) can not be cast to the dtype of the key '${n.name}' (${n.dtype}).`);
   }
 }
-class Xt {
+class Yt {
   /**
    * Constructor, optionally does copy-construction.
    * @param feeds An Array of `Feed`s, or another `FeedDict`, in which case
    *   copy-construction will be performed.
    */
   constructor(t) {
-    if (this.id2Value = {}, this.id2Mask = {}, this.name2Id = {}, t instanceof Xt)
+    if (this.id2Value = {}, this.id2Mask = {}, this.name2Id = {}, t instanceof Yt)
       for (const e in t.id2Value)
         this.id2Value[e] = t.id2Value[e], e in t.id2Mask && (this.id2Mask[e] = t.id2Mask[e]);
     else {
@@ -6551,7 +6550,7 @@ function Be(n, t, e, s) {
     h = y.sorted, p = y.recipientCounts, kn.put(c, h), xn.put(c, p);
   }
   p = {}, i || Object.assign(p, xn.get(c));
-  const f = new Xt(t);
+  const f = new Yt(t);
   for (let y = 0; y < h.length; ++y) {
     const g = h[y], m = g.sourceLayer;
     if (m instanceof nn)
@@ -6743,7 +6742,7 @@ function rt(n) {
   return ks(n);
 }
 function pi(n, t = {}) {
-  return Xe(n, qe.getMap().classNameMap, t, "constraint");
+  return Ye(n, qe.getMap().classNameMap, t, "constraint");
 }
 function ot(n) {
   if (n == null)
@@ -7088,7 +7087,7 @@ function Ur(n, t, e, s, i, r, o, a, l) {
  * =============================================================================
  */
 function Zt(n, t = {}, e = !1) {
-  return Xe(n, qe.getMap().classNameMap, t, "layer", e);
+  return Ye(n, qe.getMap().classNameMap, t, "layer", e);
 }
 /**
  * @license
@@ -7235,7 +7234,7 @@ function Un(n) {
  */
 function jr(n, t) {
   return x(() => {
-    const e = w(0.5, Ot(t)), s = Pt(Yt(t, e), n.dtype);
+    const e = w(0.5, Ot(t)), s = Pt(Xt(t, e), n.dtype);
     return nt(re(n, s), -1);
   });
 }
@@ -7245,13 +7244,13 @@ function Kr(n, t) {
 function Zf(n, t) {
   return x(() => L(W(Ze(re(n, 1), re(t, 1))), "float32"));
 }
-function Yf(n, t) {
+function Xf(n, t) {
   return x(() => L(W(Ze(re(n, 0), re(t, 1))), "float32"));
 }
-function Xf(n, t) {
+function Yf(n, t) {
   return x(() => {
-    const e = Zf(n, t), s = Yf(n, t), i = z(e, s);
-    return L(ne(Yt(i, 0), U(e, i), 0), "float32");
+    const e = Zf(n, t), s = Xf(n, t), i = z(e, s);
+    return L(ne(Xt(i, 0), U(e, i), 0), "float32");
   });
 }
 function Qf(n, t) {
@@ -7263,7 +7262,7 @@ function tm(n, t) {
 const em = Ln, nm = Ln, sm = Es, im = Es, rm = Ls, om = Ls, Hr = Ve, am = Vr, qr = vn, Sn = {
   binaryAccuracy: jr,
   categoricalAccuracy: Kr,
-  precision: Xf,
+  precision: Yf,
   categoricalCrossentropy: Hr,
   sparseCategoricalCrossentropy: qr,
   mse: em,
@@ -7838,7 +7837,7 @@ class Lt extends G {
   call(t, e) {
     return x(() => {
       t = H(t);
-      const s = new Xt();
+      const s = new Yt();
       for (let i = 0; i < this.inputs.length; ++i)
         s.add(this.inputs[i], t[i]);
       return Be(this.outputs, s, e);
@@ -8110,7 +8109,7 @@ class Lt extends G {
     const c = e.name, h = e.layers;
     for (const m of h)
       u(m);
-    for (; !Xd(o); )
+    for (; !Yd(o); )
       for (const m of h) {
         const A = r[m.name];
         if (A.name in o) {
@@ -8191,10 +8190,10 @@ function gm(n, t, e) {
   } else
     throw new Error(`The model has multiple (${s}) outputs, so ${e} must be either an array with ${s} elements or an object with ${t} keys. Provided ${e} not understood: ${JSON.stringify(n)}`);
 }
-function Yr(n, t) {
+function Xr(n, t) {
   return gm(n, t, "classWeight");
 }
-async function Xr(n, t, e, s) {
+async function Yr(n, t, e, s) {
   if (e != null) {
     const i = x(() => {
       if (n.shape.length === 1)
@@ -8314,9 +8313,9 @@ async function km(n, t, e) {
           $.batch = k, $.size = v[0].shape[0], await p.onBatchBegin(k, $);
           const R = [];
           if (e.classWeight != null) {
-            const M = Yr(e.classWeight, n.outputNames);
+            const M = Xr(e.classWeight, n.outputNames);
             for (let P = 0; P < M.length; ++P)
-              R.push(await Xr(D[P], null, M[P]));
+              R.push(await Yr(D[P], null, M[P]));
           }
           const E = v.concat(D).concat(R), F = a(E);
           J(E);
@@ -8852,7 +8851,7 @@ class Ie extends Lt {
   execute(t, e) {
     if (Array.isArray(e) && e.length === 0)
       throw new d("`outputs` is an empty Array, which is not allowed.");
-    const s = Array.isArray(e), i = s ? e : [e], r = this.retrieveSymbolicTensors(i), o = new Xt();
+    const s = Array.isArray(e), i = s ? e : [e], r = this.retrieveSymbolicTensors(i), o = new Yt();
     if (t instanceof Te && (t = [t]), Array.isArray(t)) {
       if (t.length !== this.inputs.length)
         throw new d(`The number of inputs provided (${t.length}) does not match the number of inputs of this model (${this.inputs.length}).`);
@@ -8919,7 +8918,7 @@ class Ie extends Lt {
               p.push({ key: this.inputs[y], value: h[y] });
           else
             p.push({ key: this.inputs[0], value: h });
-          const f = new Xt(p);
+          const f = new Yt(p);
           return Be(this.outputs, f);
         }).forEach((u, c) => o[c].push(u));
       return yt(o.map((a) => pe(a, 0)));
@@ -9000,10 +8999,10 @@ class Ie extends Lt {
       throw new Error("sample weight is not supported yet.");
     let u = null;
     if (i != null) {
-      const c = Yr(i, this.outputNames);
+      const c = Xr(i, this.outputNames);
       u = [];
       for (let h = 0; h < c.length; ++h)
-        u.push(await Xr(l[h], null, c[h]));
+        u.push(await Yr(l[h], null, c[h]));
     }
     return [a, l, u];
   }
@@ -9072,7 +9071,7 @@ class Ie extends Lt {
         const h = [];
         for (let g = 0; g < this.inputs.length; ++g)
           h.push({ key: this.inputs[g], value: s[g] });
-        const p = new Xt(h), f = Be(this.outputs, p, { training: !0 });
+        const p = new Yt(h), f = Be(this.outputs, p, { training: !0 });
         let y;
         for (let g = 0; g < this.lossFunctions.length; ++g) {
           const m = this.lossFunctions[g];
@@ -9110,7 +9109,7 @@ class Ie extends Lt {
       const i = t.slice(0, this.inputs.length), r = t.slice(this.inputs.length, this.inputs.length + this.outputs.length), o = [];
       for (let u = 0; u < this.inputs.length; ++u)
         o.push({ key: this.inputs[u], value: i[u] });
-      const a = new Xt(o), l = Be(this.outputs, a);
+      const a = new Yt(o), l = Be(this.outputs, a);
       for (let u = 0; u < this.lossFunctions.length; ++u) {
         const c = this.lossFunctions[u], h = nt(c(r[u], l[u]));
         u === 0 ? s = h : s = z(s, h), e.push(s);
@@ -10295,7 +10294,7 @@ function ae(n) {
   return n.getClassName();
 }
 function Kn(n, t = {}) {
-  return Xe(n, qe.getMap().classNameMap, t, "activation");
+  return Ye(n, qe.getMap().classNameMap, t, "activation");
 }
 function le(n) {
   if (n == null) {
@@ -10353,9 +10352,9 @@ function q(n) {
   return ks(n);
 }
 function Ni(n, t = {}) {
-  return Xe(n, qe.getMap().classNameMap, t, "regularizer");
+  return Ye(n, qe.getMap().classNameMap, t, "regularizer");
 }
-function Y(n) {
+function X(n) {
   if (n == null)
     return null;
   if (typeof n == "string") {
@@ -10411,7 +10410,7 @@ xo.className = "LeakyReLU";
 S(xo);
 class No extends G {
   constructor(t) {
-    if (super(t ?? {}), this.DEFAULT_ALPHA_INITIALIZER = "zeros", t == null && (t = {}), this.supportsMasking = !0, this.alphaInitializer = Z(t.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER), this.alphaRegularizer = Y(t.alphaRegularizer), this.alphaConstraint = ot(t.alphaConstraint), t.sharedAxes == null)
+    if (super(t ?? {}), this.DEFAULT_ALPHA_INITIALIZER = "zeros", t == null && (t = {}), this.supportsMasking = !0, this.alphaInitializer = Z(t.alphaInitializer || this.DEFAULT_ALPHA_INITIALIZER), this.alphaRegularizer = X(t.alphaRegularizer), this.alphaConstraint = ot(t.alphaConstraint), t.sharedAxes == null)
       this.sharedAxes = null;
     else if (Array.isArray(t.sharedAxes))
       this.sharedAxes = t.sharedAxes;
@@ -10441,7 +10440,7 @@ class No extends G {
   }
   getConfig() {
     const t = {
-      alphaInitializer: X(this.alphaInitializer),
+      alphaInitializer: Y(this.alphaInitializer),
       alphaRegularizer: q(this.alphaRegularizer),
       alphaConstraint: rt(this.alphaConstraint),
       sharedAxes: this.sharedAxes
@@ -10477,7 +10476,7 @@ class Ao extends G {
   }
   call(t, e) {
     const s = _(t);
-    return w(s, L(Yt(s, this.theta), "float32"));
+    return w(s, L(Xt(s, this.theta), "float32"));
   }
   computeOutputShape(t) {
     return t;
@@ -10619,7 +10618,7 @@ class On extends G {
   constructor(t, e) {
     if (super(e), this.bias = null, this.DEFAULT_KERNEL_INITIALIZER = "glorotNormal", this.DEFAULT_BIAS_INITIALIZER = "zeros", On.verifyArgs(e), this.rank = t, lt(this.rank, "rank"), this.rank !== 1 && this.rank !== 2 && this.rank !== 3)
       throw new B(`Convolution layer for rank other than 1, 2, or 3 (${this.rank}) is not implemented yet.`);
-    if (this.kernelSize = De(e.kernelSize, t, "kernelSize"), this.strides = De(e.strides == null ? 1 : e.strides, t, "strides"), this.padding = e.padding == null ? "valid" : e.padding, At(this.padding), this.dataFormat = e.dataFormat == null ? "channelsLast" : e.dataFormat, tt(this.dataFormat), this.activation = le(e.activation), this.useBias = e.useBias == null ? !0 : e.useBias, this.biasInitializer = Z(e.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.biasConstraint = ot(e.biasConstraint), this.biasRegularizer = Y(e.biasRegularizer), this.activityRegularizer = Y(e.activityRegularizer), this.dilationRate = De(e.dilationRate == null ? 1 : e.dilationRate, t, "dilationRate"), this.rank === 1 && Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)
+    if (this.kernelSize = De(e.kernelSize, t, "kernelSize"), this.strides = De(e.strides == null ? 1 : e.strides, t, "strides"), this.padding = e.padding == null ? "valid" : e.padding, At(this.padding), this.dataFormat = e.dataFormat == null ? "channelsLast" : e.dataFormat, tt(this.dataFormat), this.activation = le(e.activation), this.useBias = e.useBias == null ? !0 : e.useBias, this.biasInitializer = Z(e.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.biasConstraint = ot(e.biasConstraint), this.biasRegularizer = X(e.biasRegularizer), this.activityRegularizer = X(e.activityRegularizer), this.dilationRate = De(e.dilationRate == null ? 1 : e.dilationRate, t, "dilationRate"), this.rank === 1 && Array.isArray(this.dilationRate) && this.dilationRate.length !== 1)
       throw new d(`dilationRate must be a number or an array of a single number for 1D convolution, but received ${JSON.stringify(this.dilationRate)}`);
     if (this.rank === 2) {
       if (typeof this.dilationRate == "number")
@@ -10646,7 +10645,7 @@ class On extends G {
       dilationRate: this.dilationRate,
       activation: ae(this.activation),
       useBias: this.useBias,
-      biasInitializer: X(this.biasInitializer),
+      biasInitializer: Y(this.biasInitializer),
       biasRegularizer: q(this.biasRegularizer),
       activityRegularizer: q(this.activityRegularizer),
       biasConstraint: rt(this.biasConstraint)
@@ -10656,7 +10655,7 @@ class On extends G {
 }
 class Me extends On {
   constructor(t, e) {
-    super(t, e), this.kernel = null, Me.verifyArgs(e), this.filters = e.filters, lt(this.filters, "filters"), this.kernelInitializer = Z(e.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.kernelConstraint = ot(e.kernelConstraint), this.kernelRegularizer = Y(e.kernelRegularizer);
+    super(t, e), this.kernel = null, Me.verifyArgs(e), this.filters = e.filters, lt(this.filters, "filters"), this.kernelInitializer = Z(e.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.kernelConstraint = ot(e.kernelConstraint), this.kernelRegularizer = X(e.kernelRegularizer);
   }
   build(t) {
     t = j(t);
@@ -10700,7 +10699,7 @@ class Me extends On {
   getConfig() {
     const t = {
       filters: this.filters,
-      kernelInitializer: X(this.kernelInitializer),
+      kernelInitializer: Y(this.kernelInitializer),
       kernelRegularizer: q(this.kernelRegularizer),
       kernelConstraint: rt(this.kernelConstraint)
     }, e = super.getConfig();
@@ -10835,7 +10834,7 @@ class $o extends Me {
       throw new d("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");
     if (e.padding != null && e.padding !== "same" && e.padding !== "valid")
       throw new d(`SeparableConv${this.rank}D supports only padding modes: 'same' and 'valid', but received ${JSON.stringify(e.padding)}`);
-    this.depthMultiplier = e.depthMultiplier == null ? 1 : e.depthMultiplier, this.depthwiseInitializer = Z(e.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER), this.depthwiseRegularizer = Y(e.depthwiseRegularizer), this.depthwiseConstraint = ot(e.depthwiseConstraint), this.pointwiseInitializer = Z(e.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER), this.pointwiseRegularizer = Y(e.pointwiseRegularizer), this.pointwiseConstraint = ot(e.pointwiseConstraint);
+    this.depthMultiplier = e.depthMultiplier == null ? 1 : e.depthMultiplier, this.depthwiseInitializer = Z(e.depthwiseInitializer || this.DEFAULT_DEPTHWISE_INITIALIZER), this.depthwiseRegularizer = X(e.depthwiseRegularizer), this.depthwiseConstraint = ot(e.depthwiseConstraint), this.pointwiseInitializer = Z(e.depthwiseInitializer || this.DEFAULT_POINTWISE_INITIALIZER), this.pointwiseRegularizer = X(e.pointwiseRegularizer), this.pointwiseConstraint = ot(e.pointwiseConstraint);
   }
   build(t) {
     if (t = j(t), t.length < this.rank + 2)
@@ -10861,7 +10860,7 @@ class $o extends Me {
   }
   getConfig() {
     const t = super.getConfig();
-    return delete t.rank, delete t.kernelInitializer, delete t.kernelRegularizer, delete t.kernelConstraint, t.depthwiseInitializer = X(this.depthwiseInitializer), t.pointwiseInitializer = X(this.pointwiseInitializer), t.depthwiseRegularizer = q(this.depthwiseRegularizer), t.pointwiseRegularizer = q(this.pointwiseRegularizer), t.depthwiseConstraint = rt(this.depthwiseConstraint), t.pointwiseConstraint = rt(this.pointwiseConstraint), t;
+    return delete t.rank, delete t.kernelInitializer, delete t.kernelRegularizer, delete t.kernelConstraint, t.depthwiseInitializer = Y(this.depthwiseInitializer), t.pointwiseInitializer = Y(this.pointwiseInitializer), t.depthwiseRegularizer = q(this.depthwiseRegularizer), t.pointwiseRegularizer = q(this.pointwiseRegularizer), t.depthwiseConstraint = rt(this.depthwiseConstraint), t.pointwiseConstraint = rt(this.pointwiseConstraint), t;
   }
 }
 $o.className = "SeparableConv";
@@ -10985,7 +10984,7 @@ function Em(n, t, e = [1, 1], s = "valid", i, r) {
 }
 class Lo extends On {
   constructor(t) {
-    super(2, t), this.depthwiseKernel = null, this.depthMultiplier = t.depthMultiplier == null ? 1 : t.depthMultiplier, this.depthwiseInitializer = Z(t.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.depthwiseConstraint = ot(t.depthwiseConstraint), this.depthwiseRegularizer = Y(t.depthwiseRegularizer);
+    super(2, t), this.depthwiseKernel = null, this.depthMultiplier = t.depthMultiplier == null ? 1 : t.depthMultiplier, this.depthwiseInitializer = Z(t.depthwiseInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.depthwiseConstraint = ot(t.depthwiseConstraint), this.depthwiseRegularizer = X(t.depthwiseRegularizer);
   }
   build(t) {
     if (t = j(t), t.length < 4)
@@ -11015,7 +11014,7 @@ class Lo extends On {
   }
   getConfig() {
     const t = super.getConfig();
-    return t.depthMultiplier = this.depthMultiplier, t.depthwiseInitializer = X(this.depthwiseInitializer), t.depthwiseRegularizer = q(this.depthwiseRegularizer), t.depthwiseConstraint = rt(this.depthwiseRegularizer), t;
+    return t.depthMultiplier = this.depthMultiplier, t.depthwiseInitializer = Y(this.depthwiseInitializer), t.depthwiseRegularizer = q(this.depthwiseRegularizer), t.depthwiseConstraint = rt(this.depthwiseRegularizer), t;
   }
 }
 Lo.className = "DepthwiseConv2D";
@@ -11094,7 +11093,7 @@ class de extends G {
     this.states_ = t;
   }
   computeOutputShape(t) {
-    Xn(t) && (t = t[0]), t = t;
+    Yn(t) && (t = t[0]), t = t;
     let e = this.cell.stateSize;
     Array.isArray(e) || (e = [e]);
     const s = e[0];
@@ -11139,7 +11138,7 @@ class de extends G {
   build(t) {
     if (this.numConstants != null)
       throw new B("Constants support is not implemented in RNN yet.");
-    Xn(t) && (t = t[0]), t = t;
+    Yn(t) && (t = t[0]), t = t;
     const e = this.stateful ? t[0] : null, s = t.slice(2);
     this.inputSpec[0] = new it({ shape: [e, null, ...s] });
     const i = [t[0]].concat(t.slice(2));
@@ -11275,7 +11274,7 @@ class _n extends G {
 }
 class Ms extends _n {
   constructor(t) {
-    super(t), this.DEFAULT_ACTIVATION = "tanh", this.DEFAULT_KERNEL_INITIALIZER = "glorotNormal", this.DEFAULT_RECURRENT_INITIALIZER = "orthogonal", this.DEFAULT_BIAS_INITIALIZER = "zeros", this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation == null ? this.DEFAULT_ACTIVATION : t.activation), this.useBias = t.useBias == null ? !0 : t.useBias, this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.recurrentInitializer = Z(t.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.kernelRegularizer = Y(t.kernelRegularizer), this.recurrentRegularizer = Y(t.recurrentRegularizer), this.biasRegularizer = Y(t.biasRegularizer), this.kernelConstraint = ot(t.kernelConstraint), this.recurrentConstraint = ot(t.recurrentConstraint), this.biasConstraint = ot(t.biasConstraint), this.dropout = Fe([1, oe([0, t.dropout == null ? 0 : t.dropout])]), this.recurrentDropout = Fe([
+    super(t), this.DEFAULT_ACTIVATION = "tanh", this.DEFAULT_KERNEL_INITIALIZER = "glorotNormal", this.DEFAULT_RECURRENT_INITIALIZER = "orthogonal", this.DEFAULT_BIAS_INITIALIZER = "zeros", this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation == null ? this.DEFAULT_ACTIVATION : t.activation), this.useBias = t.useBias == null ? !0 : t.useBias, this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.recurrentInitializer = Z(t.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.kernelRegularizer = X(t.kernelRegularizer), this.recurrentRegularizer = X(t.recurrentRegularizer), this.biasRegularizer = X(t.biasRegularizer), this.kernelConstraint = ot(t.kernelConstraint), this.recurrentConstraint = ot(t.recurrentConstraint), this.biasConstraint = ot(t.biasConstraint), this.dropout = Fe([1, oe([0, t.dropout == null ? 0 : t.dropout])]), this.recurrentDropout = Fe([
       1,
       oe([0, t.recurrentDropout == null ? 0 : t.recurrentDropout])
     ]), this.dropoutFunc = t.dropoutFunc, this.stateSize = this.units, this.dropoutMask = null, this.recurrentDropoutMask = null;
@@ -11319,9 +11318,9 @@ class Ms extends _n {
       units: this.units,
       activation: ae(this.activation),
       useBias: this.useBias,
-      kernelInitializer: X(this.kernelInitializer),
-      recurrentInitializer: X(this.recurrentInitializer),
-      biasInitializer: X(this.biasInitializer),
+      kernelInitializer: Y(this.kernelInitializer),
+      recurrentInitializer: Y(this.recurrentInitializer),
+      biasInitializer: Y(this.biasInitializer),
       kernelRegularizer: q(this.kernelRegularizer),
       recurrentRegularizer: q(this.recurrentRegularizer),
       biasRegularizer: q(this.biasRegularizer),
@@ -11359,7 +11358,7 @@ class _s extends _n {
   constructor(t) {
     if (super(t), this.DEFAULT_ACTIVATION = "tanh", this.DEFAULT_RECURRENT_ACTIVATION = "hardSigmoid", this.DEFAULT_KERNEL_INITIALIZER = "glorotNormal", this.DEFAULT_RECURRENT_INITIALIZER = "orthogonal", this.DEFAULT_BIAS_INITIALIZER = "zeros", t.resetAfter)
       throw new d("GRUCell does not support reset_after parameter set to true.");
-    this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation === void 0 ? this.DEFAULT_ACTIVATION : t.activation), this.recurrentActivation = le(t.recurrentActivation === void 0 ? this.DEFAULT_RECURRENT_ACTIVATION : t.recurrentActivation), this.useBias = t.useBias == null ? !0 : t.useBias, this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.recurrentInitializer = Z(t.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.kernelRegularizer = Y(t.kernelRegularizer), this.recurrentRegularizer = Y(t.recurrentRegularizer), this.biasRegularizer = Y(t.biasRegularizer), this.kernelConstraint = ot(t.kernelConstraint), this.recurrentConstraint = ot(t.recurrentConstraint), this.biasConstraint = ot(t.biasConstraint), this.dropout = Fe([1, oe([0, t.dropout == null ? 0 : t.dropout])]), this.recurrentDropout = Fe([
+    this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation === void 0 ? this.DEFAULT_ACTIVATION : t.activation), this.recurrentActivation = le(t.recurrentActivation === void 0 ? this.DEFAULT_RECURRENT_ACTIVATION : t.recurrentActivation), this.useBias = t.useBias == null ? !0 : t.useBias, this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.recurrentInitializer = Z(t.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.kernelRegularizer = X(t.kernelRegularizer), this.recurrentRegularizer = X(t.recurrentRegularizer), this.biasRegularizer = X(t.biasRegularizer), this.kernelConstraint = ot(t.kernelConstraint), this.recurrentConstraint = ot(t.recurrentConstraint), this.biasConstraint = ot(t.biasConstraint), this.dropout = Fe([1, oe([0, t.dropout == null ? 0 : t.dropout])]), this.recurrentDropout = Fe([
       1,
       oe([0, t.recurrentDropout == null ? 0 : t.recurrentDropout])
     ]), this.dropoutFunc = t.dropoutFunc, this.implementation = t.implementation, this.stateSize = this.units, this.dropoutMask = null, this.recurrentDropoutMask = null;
@@ -11407,9 +11406,9 @@ class _s extends _n {
       activation: ae(this.activation),
       recurrentActivation: ae(this.recurrentActivation),
       useBias: this.useBias,
-      kernelInitializer: X(this.kernelInitializer),
-      recurrentInitializer: X(this.recurrentInitializer),
-      biasInitializer: X(this.biasInitializer),
+      kernelInitializer: Y(this.kernelInitializer),
+      recurrentInitializer: Y(this.recurrentInitializer),
+      biasInitializer: Y(this.biasInitializer),
       kernelRegularizer: q(this.kernelRegularizer),
       recurrentRegularizer: q(this.recurrentRegularizer),
       biasRegularizer: q(this.biasRegularizer),
@@ -11447,7 +11446,7 @@ _o.className = "GRU";
 S(_o);
 class Rn extends _n {
   constructor(t) {
-    super(t), this.DEFAULT_ACTIVATION = "tanh", this.DEFAULT_RECURRENT_ACTIVATION = "hardSigmoid", this.DEFAULT_KERNEL_INITIALIZER = "glorotNormal", this.DEFAULT_RECURRENT_INITIALIZER = "orthogonal", this.DEFAULT_BIAS_INITIALIZER = "zeros", this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation === void 0 ? this.DEFAULT_ACTIVATION : t.activation), this.recurrentActivation = le(t.recurrentActivation === void 0 ? this.DEFAULT_RECURRENT_ACTIVATION : t.recurrentActivation), this.useBias = t.useBias == null ? !0 : t.useBias, this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.recurrentInitializer = Z(t.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.unitForgetBias = t.unitForgetBias, this.kernelRegularizer = Y(t.kernelRegularizer), this.recurrentRegularizer = Y(t.recurrentRegularizer), this.biasRegularizer = Y(t.biasRegularizer), this.kernelConstraint = ot(t.kernelConstraint), this.recurrentConstraint = ot(t.recurrentConstraint), this.biasConstraint = ot(t.biasConstraint), this.dropout = Fe([1, oe([0, t.dropout == null ? 0 : t.dropout])]), this.recurrentDropout = Fe([
+    super(t), this.DEFAULT_ACTIVATION = "tanh", this.DEFAULT_RECURRENT_ACTIVATION = "hardSigmoid", this.DEFAULT_KERNEL_INITIALIZER = "glorotNormal", this.DEFAULT_RECURRENT_INITIALIZER = "orthogonal", this.DEFAULT_BIAS_INITIALIZER = "zeros", this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation === void 0 ? this.DEFAULT_ACTIVATION : t.activation), this.recurrentActivation = le(t.recurrentActivation === void 0 ? this.DEFAULT_RECURRENT_ACTIVATION : t.recurrentActivation), this.useBias = t.useBias == null ? !0 : t.useBias, this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.recurrentInitializer = Z(t.recurrentInitializer || this.DEFAULT_RECURRENT_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.unitForgetBias = t.unitForgetBias, this.kernelRegularizer = X(t.kernelRegularizer), this.recurrentRegularizer = X(t.recurrentRegularizer), this.biasRegularizer = X(t.biasRegularizer), this.kernelConstraint = ot(t.kernelConstraint), this.recurrentConstraint = ot(t.recurrentConstraint), this.biasConstraint = ot(t.biasConstraint), this.dropout = Fe([1, oe([0, t.dropout == null ? 0 : t.dropout])]), this.recurrentDropout = Fe([
       1,
       oe([0, t.recurrentDropout == null ? 0 : t.recurrentDropout])
     ]), this.dropoutFunc = t.dropoutFunc, this.implementation = t.implementation, this.stateSize = [this.units, this.units], this.dropoutMask = null, this.recurrentDropoutMask = null;
@@ -11512,9 +11511,9 @@ class Rn extends _n {
       activation: ae(this.activation),
       recurrentActivation: ae(this.recurrentActivation),
       useBias: this.useBias,
-      kernelInitializer: X(this.kernelInitializer),
-      recurrentInitializer: X(this.recurrentInitializer),
-      biasInitializer: X(this.biasInitializer),
+      kernelInitializer: Y(this.kernelInitializer),
+      recurrentInitializer: Y(this.recurrentInitializer),
+      biasInitializer: Y(this.biasInitializer),
       unitForgetBias: this.unitForgetBias,
       kernelRegularizer: q(this.kernelRegularizer),
       recurrentRegularizer: q(this.recurrentRegularizer),
@@ -11581,7 +11580,7 @@ class Rs extends _n {
     });
   }
   build(t) {
-    Xn(t) && (t = t[0]), t = t;
+    Yn(t) && (t = t[0]), t = t;
     let e;
     this.cells.forEach((s, i) => {
       ge(`RNNCell_${i}`, () => {
@@ -11891,7 +11890,7 @@ class Po extends G {
       let e = null;
       t.batchSize != null && (e = t.batchSize), this.batchInputShape = [e, t.inputDim];
     }
-    this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation), t.useBias != null && (this.useBias = t.useBias), this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.kernelConstraint = ot(t.kernelConstraint), this.biasConstraint = ot(t.biasConstraint), this.kernelRegularizer = Y(t.kernelRegularizer), this.biasRegularizer = Y(t.biasRegularizer), this.activityRegularizer = Y(t.activityRegularizer), this.supportsMasking = !0, this.inputSpec = [{ minNDim: 2 }];
+    this.units = t.units, lt(this.units, "units"), this.activation = le(t.activation), t.useBias != null && (this.useBias = t.useBias), this.kernelInitializer = Z(t.kernelInitializer || this.DEFAULT_KERNEL_INITIALIZER), this.biasInitializer = Z(t.biasInitializer || this.DEFAULT_BIAS_INITIALIZER), this.kernelConstraint = ot(t.kernelConstraint), this.biasConstraint = ot(t.biasConstraint), this.kernelRegularizer = X(t.kernelRegularizer), this.biasRegularizer = X(t.biasRegularizer), this.activityRegularizer = X(t.activityRegularizer), this.supportsMasking = !0, this.inputSpec = [{ minNDim: 2 }];
   }
   build(t) {
     t = j(t);
@@ -11916,8 +11915,8 @@ class Po extends G {
       units: this.units,
       activation: ae(this.activation),
       useBias: this.useBias,
-      kernelInitializer: X(this.kernelInitializer),
-      biasInitializer: X(this.biasInitializer),
+      kernelInitializer: Y(this.kernelInitializer),
+      biasInitializer: Y(this.biasInitializer),
       kernelRegularizer: q(this.kernelRegularizer),
       biasRegularizer: q(this.biasRegularizer),
       activityRegularizer: q(this.activityRegularizer),
@@ -12139,7 +12138,7 @@ class Jo extends G {
       let e = null;
       t.batchSize != null && (e = t.batchSize), t.inputLength == null ? this.batchInputShape = [e, null] : this.batchInputShape = [e].concat(H(t.inputLength));
     }
-    this.inputDim = t.inputDim, lt(this.inputDim, "inputDim"), this.outputDim = t.outputDim, lt(this.outputDim, "outputDim"), this.embeddingsInitializer = Z(t.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER), this.embeddingsRegularizer = Y(t.embeddingsRegularizer), this.activityRegularizer = Y(t.activityRegularizer), this.embeddingsConstraint = ot(t.embeddingsConstraint), this.maskZero = t.maskZero, this.supportsMasking = t.maskZero, this.inputLength = t.inputLength;
+    this.inputDim = t.inputDim, lt(this.inputDim, "inputDim"), this.outputDim = t.outputDim, lt(this.outputDim, "outputDim"), this.embeddingsInitializer = Z(t.embeddingsInitializer || this.DEFAULT_EMBEDDINGS_INITIALIZER), this.embeddingsRegularizer = X(t.embeddingsRegularizer), this.activityRegularizer = X(t.activityRegularizer), this.embeddingsConstraint = ot(t.embeddingsConstraint), this.maskZero = t.maskZero, this.supportsMasking = t.maskZero, this.inputLength = t.inputLength;
   }
   build(t) {
     this.embeddings = this.addWeight("embeddings", [this.inputDim, this.outputDim], this.dtype, this.embeddingsInitializer, this.embeddingsRegularizer, !0, this.embeddingsConstraint), this.built = !0;
@@ -12181,7 +12180,7 @@ class Jo extends G {
     const t = {
       inputDim: this.inputDim,
       outputDim: this.outputDim,
-      embeddingsInitializer: X(this.embeddingsInitializer),
+      embeddingsInitializer: Y(this.embeddingsInitializer),
       embeddingsRegularizer: q(this.embeddingsRegularizer),
       activityRegularizer: q(this.activityRegularizer),
       embeddingsConstraint: rt(this.embeddingsConstraint),
@@ -12355,7 +12354,7 @@ class Zo extends ve {
 }
 Zo.className = "Add";
 S(Zo);
-class Yo extends ve {
+class Xo extends ve {
   constructor(t) {
     super(t);
   }
@@ -12368,9 +12367,9 @@ class Yo extends ve {
     });
   }
 }
-Yo.className = "Multiply";
-S(Yo);
-class Xo extends ve {
+Xo.className = "Multiply";
+S(Xo);
+class Yo extends ve {
   constructor(t) {
     super(t);
   }
@@ -12383,8 +12382,8 @@ class Xo extends ve {
     });
   }
 }
-Xo.className = "Average";
-S(Xo);
+Yo.className = "Average";
+S(Yo);
 class Qo extends ve {
   constructor(t) {
     super(t);
@@ -12669,7 +12668,7 @@ class ra extends G {
         const s = this._getNoiseShape(t);
         return en(() => {
           const r = _(t), a = -1.6732632423543772 * 1.0507009873554805;
-          let l = Ye($n(s), this.rate);
+          let l = Xe($n(s), this.rate);
           l = Pt(l, "float32");
           const u = ((1 - this.rate) * (1 + this.rate * a ** 2)) ** -0.5, c = -u * a * this.rate, h = z(w(r, l), w(z(l, -1), a));
           return z(w(h, u), c);
@@ -12722,7 +12721,7 @@ function _m(n, t, e, s, i = 1e-3) {
 }
 class oa extends G {
   constructor(t) {
-    t == null && (t = {}), super(t), this.supportsMasking = !0, this.axis = t.axis == null ? -1 : t.axis, this.momentum = t.momentum == null ? 0.99 : t.momentum, this.epsilon = t.epsilon == null ? 1e-3 : t.epsilon, this.center = t.center == null ? !0 : t.center, this.scale = t.scale == null ? !0 : t.scale, this.betaInitializer = Z(t.betaInitializer || "zeros"), this.gammaInitializer = Z(t.gammaInitializer || "ones"), this.movingMeanInitializer = Z(t.movingMeanInitializer || "zeros"), this.movingVarianceInitializer = Z(t.movingVarianceInitializer || "ones"), this.betaConstraint = ot(t.betaConstraint), this.gammaConstraint = ot(t.gammaConstraint), this.betaRegularizer = Y(t.betaRegularizer), this.gammaRegularizer = Y(t.gammaRegularizer);
+    t == null && (t = {}), super(t), this.supportsMasking = !0, this.axis = t.axis == null ? -1 : t.axis, this.momentum = t.momentum == null ? 0.99 : t.momentum, this.epsilon = t.epsilon == null ? 1e-3 : t.epsilon, this.center = t.center == null ? !0 : t.center, this.scale = t.scale == null ? !0 : t.scale, this.betaInitializer = Z(t.betaInitializer || "zeros"), this.gammaInitializer = Z(t.gammaInitializer || "ones"), this.movingMeanInitializer = Z(t.movingMeanInitializer || "zeros"), this.movingVarianceInitializer = Z(t.movingVarianceInitializer || "ones"), this.betaConstraint = ot(t.betaConstraint), this.gammaConstraint = ot(t.gammaConstraint), this.betaRegularizer = X(t.betaRegularizer), this.gammaRegularizer = X(t.gammaRegularizer);
   }
   build(t) {
     t = j(t);
@@ -12768,10 +12767,10 @@ class oa extends G {
       epsilon: this.epsilon,
       center: this.center,
       scale: this.scale,
-      betaInitializer: X(this.betaInitializer),
-      gammaInitializer: X(this.gammaInitializer),
-      movingMeanInitializer: X(this.movingMeanInitializer),
-      movingVarianceInitializer: X(this.movingVarianceInitializer),
+      betaInitializer: Y(this.betaInitializer),
+      gammaInitializer: Y(this.gammaInitializer),
+      movingMeanInitializer: Y(this.movingMeanInitializer),
+      movingVarianceInitializer: Y(this.movingVarianceInitializer),
       betaRegularizer: q(this.betaRegularizer),
       gammaRegularizer: q(this.gammaRegularizer),
       betaConstraint: rt(this.betaConstraint),
@@ -12793,7 +12792,7 @@ class aa extends G {
           throw new Error(`Expected axis to be an array of integers, but received ${JSON.stringify(this.axis)}`);
     } else
       throw new Error(`Expected axis to be an integer or an array of integers, but received ${JSON.stringify(this.axis)}`);
-    this.epsilon = t.epsilon == null ? 1e-3 : t.epsilon, this.center = t.center == null ? !0 : t.center, this.scale = t.scale == null ? !0 : t.scale, this.betaInitializer = Z(t.betaInitializer || "zeros"), this.gammaInitializer = Z(t.gammaInitializer || "ones"), this.betaRegularizer = Y(t.betaRegularizer), this.gammaRegularizer = Y(t.gammaRegularizer), this.supportsMasking = !0;
+    this.epsilon = t.epsilon == null ? 1e-3 : t.epsilon, this.center = t.center == null ? !0 : t.center, this.scale = t.scale == null ? !0 : t.scale, this.betaInitializer = Z(t.betaInitializer || "zeros"), this.gammaInitializer = Z(t.gammaInitializer || "ones"), this.betaRegularizer = X(t.betaRegularizer), this.gammaRegularizer = X(t.gammaRegularizer), this.supportsMasking = !0;
   }
   build(t) {
     t = j(t);
@@ -12830,8 +12829,8 @@ class aa extends G {
       epsilon: this.epsilon,
       center: this.center,
       scale: this.scale,
-      betaInitializer: X(this.betaInitializer),
-      gammaInitializer: X(this.gammaInitializer),
+      betaInitializer: Y(this.betaInitializer),
+      gammaInitializer: Y(this.gammaInitializer),
       betaRegularizer: q(this.betaRegularizer),
       gammaRegularizer: q(this.gammaRegularizer)
     }, e = super.getConfig();
@@ -13438,7 +13437,7 @@ class $a extends G {
       t.rank === 3 ? (c = !0, u = Dn([t])) : u = t;
       for (let D = 0; D < u.shape[0]; D++)
         m.push(g);
-      const A = ec(m, [m.length, 4]), k = nc(0, m.length, 1, "int32"), v = Pm(u, A, k, [i, r], "nearest");
+      const A = nc(m, [m.length, 4]), k = sc(0, m.length, 1, "int32"), v = Pm(u, A, k, [i, r], "nearest");
       return Pt(c ? _(dn(v)) : v, l);
     });
   }
@@ -13529,7 +13528,7 @@ class Ta extends G {
               Received countWeights=${e.countWeights}`);
         s = _(e.countWeights);
       }
-      const i = Ee(t), r = sc(t), o = Yt(this.numTokens, i).bufferSync().get(0), a = Ye(r, 0).bufferSync().get(0);
+      const i = Ee(t), r = Ju(t), o = Xt(this.numTokens, i).bufferSync().get(0), a = Xe(r, 0).bufferSync().get(0);
       if (!(o && a))
         throw new d(`Input values must be between 0 < values <= numTokens with numTokens=${this.numTokens}`);
       return Um(t, this.outputMode, this.numTokens, s);