@genai-fi/nanogpt 0.16.1 → 0.17.1

package/dist/Generator.js

@@ -1,23 +1,23 @@
 import { E as Ui } from "./index-DvYrXKkX.js";
-import { o as Hi, q as Xi, E as Ki, dl as Ss, a5 as pe, ab as _, as as oo, at as ao, h as Oe, aY as Dt, ay as ro, az as io, au as ji, av as Ft, aD as Ge, ae as co, aG as ws, aH as qi, U as G, af as _e, aI as Yi, H as Ns, aJ as Rs, R as Qi, aj as Zi, x as te, D as lo, _ as Ne, a9 as ee, bb as uo, c7 as Ts, c8 as Es, cO as po, bm as ho, ah as ue, Q as Ye, bn as fo, bo as mo, c9 as go, ca as Ds, cb as Fs, cc as Ps, cd as Os, ce as As, bp as xo, ac as nt, cy as Co, cP as bo, cQ as Io, bs as yo, br as ko, bd as $o, de as vo, C as _s, cS as So, ao as wo, z as No, bt as Ro, c2 as $e, cD as To, bu as Eo, cz as Do, cT as Fo, cA as Po, bv as Ls, bw as Vs, bf as Oo, bx as Ao, am as Qe, V as Ji, by as _o, cB as Lo, cf as Vo, bz as Wo, cF as Mo, cG as Bo, df as Go, cg as zo, bX as ns, bR as St, bU as Ws, cU as yn, dt as ut, du as Uo, cV as kn, dg as ec, N as tc, be as Ho, cW as Xo, bB as Ms, A as Ko, aX as jo, aV as mt, cp as qo, dc as Yo, dd as Qo, bg as Zo, cE as Jo, di as ea, al as ta, G as sa, cq as na, ch as Bs, ci as Gs, cj as zs, dj as oa, b3 as Us, b4 as Hs, bD as Xs, cl as Ks, ck as aa, cY as ra, aM as sc, bE as ia, cC as ca, cZ as la, c_ as ua, dk as da, aP as pa, aZ as ha, cm as fa, bQ as ma, M as js, I as ga, bi as xa, bk as Ca, bj as ba, bF as Ia, d3 as ya, bG as ka, P as $a, a6 as va, bH as Sa, c$ as qs, dv as wa, dw as Na, dx as Ra, Z as Ta, cn as Ys, b9 as Ea, d0 as Da, ba as Fa, d1 as Pa, bJ as Oa, bh as Aa, b6 as Qs, ak as _a, dm as La, aL as Va, bL as Zs, co as Js, bM as en, bN as tn, bC as sn, bI as Wa, dy as Ma, dz as Ba, dn as Ga, dp as za, dq as Ua, J as Ha, d2 as Xa, aK as nn, cr as Ka, dr as ja, dA as qa, dB as Ya, cs as on, bq as an, ds as Qa, T as Za, ct as Ja, bl as er, dC as rn, cu as tr, b8 as sr, bO as nr, c as or, dD as ar, dE as $n, aw as vn, ax as nc, t as rr, a as oc, dF as ac, dG as rc, c0 as ic, ar as cc, bP as lc, bV as uc, S as dc, bW as pc, aQ as hc, aq as fc, bS as mc, bT as gc, bY as xc, aS as ir, bA as Cc, aN as bc, bZ as Ic, F as yc, b_ as kc, dh as $c, a$ as vc, b0 as Sc, b1 as wc, b2 as Nc, aO as Rc, b$ as Tc, b5 as Ec, c5 as Dc, ap as Fc, c1 as Pc, aW as cr, bK as Oc, aF as Ac, c3 as _c, b7 as Lc, c4 as Vc, k as Wc } from "./index-CUXkjxiT.js";
+import { o as Hi, q as Xi, E as Ki, dl as Ss, a5 as pe, ab as _, as as oo, at as ao, h as Oe, aY as Dt, ay as ro, az as io, au as ji, av as Ft, aD as Ge, ae as co, aG as ws, aH as qi, U as G, af as _e, aI as Yi, H as Ns, aJ as Rs, R as Qi, aj as Ji, x as te, D as lo, _ as Ne, a9 as ee, bb as uo, c7 as Ts, c8 as Es, cO as po, bm as ho, ah as ue, Q as Ye, bn as fo, bo as mo, c9 as go, ca as Ds, cb as Fs, cc as Ps, cd as Os, ce as As, bp as xo, ac as nt, cy as Co, cP as bo, cQ as Io, bs as yo, br as ko, bd as $o, de as vo, C as _s, cS as So, ao as wo, z as No, bt as Ro, c2 as $e, cD as To, bu as Eo, cz as Do, cT as Fo, cA as Po, bv as Ls, bw as Vs, bf as Oo, bx as Ao, am as Qe, V as Zi, by as _o, cB as Lo, cf as Vo, bz as Wo, cF as Mo, cG as Bo, df as Go, cg as zo, bX as ns, bR as St, bU as Ws, cU as yn, dt as ut, du as Uo, cV as kn, dg as ec, N as tc, be as Ho, cW as Xo, bB as Ms, A as Ko, aX as jo, aV as mt, cp as qo, dc as Yo, dd as Qo, bg as Jo, cE as Zo, di as ea, al as ta, G as sa, cq as na, ch as Bs, ci as Gs, cj as zs, dj as oa, b3 as Us, b4 as Hs, bD as Xs, cl as Ks, ck as aa, cY as ra, aM as sc, bE as ia, cC as ca, cZ as la, c_ as ua, dk as da, aP as pa, aZ as ha, cm as fa, bQ as ma, M as js, I as ga, bi as xa, bk as Ca, bj as ba, bF as Ia, d3 as ya, bG as ka, P as $a, a6 as va, bH as Sa, c$ as qs, dv as wa, dw as Na, dx as Ra, Z as Ta, cn as Ys, b9 as Ea, d0 as Da, ba as Fa, d1 as Pa, bJ as Oa, bh as Aa, b6 as Qs, ak as _a, dm as La, aL as Va, bL as Js, co as Zs, bM as en, bN as tn, bC as sn, bI as Wa, dy as Ma, dz as Ba, dn as Ga, dp as za, dq as Ua, J as Ha, d2 as Xa, aK as nn, cr as Ka, dr as ja, dA as qa, dB as Ya, cs as on, bq as an, ds as Qa, T as Ja, ct as Za, bl as er, dC as rn, cu as tr, b8 as sr, bO as nr, c as or, dD as ar, dE as $n, aw as vn, ax as nc, t as rr, a as oc, dF as ac, dG as rc, c0 as ic, ar as cc, bP as lc, bV as uc, S as dc, bW as pc, aQ as hc, aq as fc, bS as mc, bT as gc, bY as xc, aS as ir, bA as Cc, aN as bc, bZ as Ic, F as yc, b_ as kc, dh as $c, a$ as vc, b0 as Sc, b1 as wc, b2 as Nc, aO as Rc, b$ as Tc, b5 as Ec, c5 as Dc, ap as Fc, c1 as Pc, aW as cr, bK as Oc, aF as Ac, c3 as _c, b7 as Lc, c4 as Vc, k as Wc } from "./index-CUXkjxiT.js";
 import { n as Mc } from "./random_width-DN5ZtQkM.js";
 import { t as Bc } from "./zeros_like-CWjDdwr-.js";
 import "./index-Cp39cXWe.js";
 import "./dataset-CGGp1z9P.js";
-import { a as j, u as ae, c as ot, i as at, b as Gc, d as wt, t as Re, e as gt, f as dt, g as lr, r as Nt, h as Ae, j as zc, k as Uc, l as cn, z as Hc, m as ln, n as ur, o as Xc, p as Kc, q as jc, v as qc, w as Yc, x as Qc, y as Zc, A as Jc, B as el, C as tl, D as lt, E as sl, F as nl, G as dr, H as ol, I as al, J as rl, K as il, L as cl, M as ll, N as ul, O as dl, P as pl, Q as hl, R as fl, S as ml, T as gl, U as xl, V as Cl, W as bl, X as Il, Y as yl, Z as kl, _ as $l, $ as vl, a0 as Sl, a1 as wl, a2 as Nl, a3 as Rl, a4 as Tl, a5 as El, a6 as Dl, a7 as Fl, a8 as Pl, a9 as Ol, aa as Al, ab as _l, ac as Ll, ad as Vl, ae as Wl, af as Ml, ag as Bl, ah as Gl, ai as zl } from "./shared-CagdqkLh.js";
+import { a as j, u as ae, c as ot, i as at, b as Gc, d as wt, t as Re, e as gt, f as dt, g as lr, r as Nt, h as Ae, j as zc, k as Uc, l as cn, z as Hc, m as ln, n as ur, o as Xc, p as Kc, q as jc, v as qc, w as Yc, x as Qc, y as Jc, A as Zc, B as el, C as tl, D as lt, E as sl, F as nl, G as dr, H as ol, I as al, J as rl, K as il, L as cl, M as ll, N as ul, O as dl, P as pl, Q as hl, R as fl, S as ml, T as gl, U as xl, V as Cl, W as bl, X as Il, Y as yl, Z as kl, _ as $l, $ as vl, a0 as Sl, a1 as wl, a2 as Nl, a3 as Rl, a4 as Tl, a5 as El, a6 as Dl, a7 as Fl, a8 as Pl, a9 as Ol, aa as Al, ab as _l, ac as Ll, ad as Vl, ae as Wl, af as Ml, ag as Bl, ah as Gl, ai as zl } from "./shared-CagdqkLh.js";
 import { m as pt, g as pr, s as Ul, c as Hl, b as Xl, d as Kl, a as jl, e as ql } from "./complex_util-Yc1A_gV1.js";
 import { a as ge, b as xe, d as ke, c as ve, e as Te, g as os } from "./axis_util-GTVlo58H.js";
-import { k as Ze, h as Le, i as Je, j as rt, b as Se, d as xt, g as as } from "./step-BTxPtq1r.js";
+import { k as Je, h as Le, i as Ze, j as rt, b as Se, d as xt, g as as } from "./step-BTxPtq1r.js";
 import { z as rs, A as is, B as cs, C as hr, D as fr, F as mr, G as gr, H as xr, I as Cr, J as br, y as Ir, x as yr, w as kr, u as $r, t as vr, E as Sr, K as wr, L as Nr, M as Rr, c as Tr, f as Yl } from "./backend_util-GaFarB78.js";
 import { a as Er, c as Ue } from "./concat_util-D18dJ4fD.js";
-import { c as Dr, a as Ql, s as Zl } from "./segment_util-Dasb2Zaf.js";
-import { s as Jl } from "./index-CieiGp4Y.js";
+import { c as Dr, a as Ql, s as Jl } from "./segment_util-Dasb2Zaf.js";
+import { s as Zl } from "./index-CieiGp4Y.js";
 import { n as Fr, b as Pr, a as Or } from "./non_max_suppression_impl-B2W7YjZB.js";
 import { c as Ct } from "./scatter_nd_util-ByNJaL6I.js";
 import { S as Ar, a as _r } from "./selu_util-BLhIqRkw.js";
 import { s as Lr, b as Vr, p as eu, a as tu, i as su, c as nu } from "./slice_util-CC35pLmT.js";
 import { h as Sn, j as ou, k as au, l as ru, m as iu, n as cu, o as lu, P as un, p as Ve, u as Pe, q as Wr, c as Mr, T as De, E as Br, g as Gr, a as zr, r as uu, s as du, t as Y, v as pu, w as wn, x as hu, y as fu, z as Pt, A as Ot, B as mu, C as gu, D as bs, F as Gt, G as zt, H as xu, I as Cu, J as Nn, K as bu, L as Iu, M as fs, N as yu, O as ku, Q as $u, R as Ut, S as ms, U as vu, f as he, V as be, W as Ht, X as Xt, Y as Su, d as Rn, e as Tn, i as Ur, Z as wu, _ as Nu, $ as Ru, a0 as Tu, a1 as Eu, a2 as Du, a3 as At } from "./gpgpu_math-BwvV12df.js";
-import { s as Hr, a as Fu, t as Xr, b as Pu, c as Ou, d as Kr, e as Au, n as _u, f as Lu, g as Vu, h as Wu, i as Mu, j as Bu, k as Gu, l as zu, o as Uu, p as Hu, q as Xu, r as Ku, u as ju, v as qu, w as Yu, x as Qu, y as Zu, z as Ju, A as ed, B as td, C as sd, D as nd, E as od, F as ad, G as rd, H as id, I as cd, J as ld, K as ud, L as dd, M as jr, N as pd, O as hd, P as fd, Q as md, R as gd, S as xd, T as Cd, U as bd, V as Id, W as yd } from "./shared-3agzAqQ_.js";
+import { s as Hr, a as Fu, t as Xr, b as Pu, c as Ou, d as Kr, e as Au, n as _u, f as Lu, g as Vu, h as Wu, i as Mu, j as Bu, k as Gu, l as zu, o as Uu, p as Hu, q as Xu, r as Ku, u as ju, v as qu, w as Yu, x as Qu, y as Ju, z as Zu, A as ed, B as td, C as sd, D as nd, E as od, F as ad, G as rd, H as id, I as cd, J as ld, K as ud, L as dd, M as jr, N as pd, O as hd, P as fd, Q as md, R as gd, S as xd, T as Cd, U as bd, V as Id, W as yd } from "./shared-3agzAqQ_.js";
 import { a as ye, c as kd, U as st, d as qe, e as ze, A as En, f as bt, B as dn, h as pn, m as Rt, u as se, C as We, b as Ce, i as Fe, j as hn, k as it, l as It, n as $d, o as vd, p as Sd, q as wd } from "./kernel_funcs_utils-pq0CK9co.js";
 import { R as Nd, r as U, a as Rd } from "./Reshape-BW__R4mZ.js";
 import { M as qr } from "./matMulGelu-JNLZqKQp.js";
@@ -242,12 +242,12 @@ co(
   1
   /* priority */
 );
-const Zr = ae(ws, (a) => a >= 0 ? a : Math.exp(a) - 1), Xd = {
+const Jr = ae(ws, (a) => a >= 0 ? a : Math.exp(a) - 1), Xd = {
   kernelName: ws,
   backendName: "cpu",
-  kernelFunc: Zr
+  kernelFunc: Jr
 };
-function Jr(a) {
+function Zr(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { alpha: o } = n;
   j([s], "leakyRelu");
   const r = G(s.shape), c = e.data.get(s.dataId).values, i = _e("float32", r);
@@ -258,7 +258,7 @@ function Jr(a) {
 const Kd = {
   kernelName: qi,
   backendName: "cpu",
-  kernelFunc: Jr
+  kernelFunc: Zr
 };
 const jd = ot((a, t) => a < 0 ? t * a : a);
 function ei(a) {
@@ -288,19 +288,19 @@ function Qt(a, t, e, n, s) {
   if (e === "relu")
     return ti({ inputs: { x: t }, backend: a });
   if (e === "elu")
-    return Zr({ inputs: { x: t }, backend: a });
+    return Jr({ inputs: { x: t }, backend: a });
   if (e === "relu6")
     return si({ inputs: { x: t }, backend: a });
   if (e === "prelu")
     return ei({ inputs: { x: t, alpha: n }, backend: a });
   if (e === "leakyrelu")
-    return Jr({ inputs: { x: t }, backend: a, attrs: { alpha: s } });
+    return Zr({ inputs: { x: t }, backend: a, attrs: { alpha: s } });
   if (e === "sigmoid")
     return Gc({ inputs: { x: t }, backend: a });
   throw new Error(`Activation ${e} has not been implemented for the CPU backend.`);
 }
 function ce(a) {
-  const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { shape: o } = n, r = G(s.shape), c = Zi(o, r), i = G(c);
+  const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { shape: o } = n, r = G(s.shape), c = Ji(o, r), i = G(c);
   te(r === i, () => `The new shape (${c}) has ${i} elements and the old shape (${s.shape}) has ${r} elements. The new shape and old shape must have the same number of elements.`), e.incRef(s.dataId);
   const l = e.data.get(s.dataId);
   if (l.complexTensorInfos != null) {
@@ -309,7 +309,7 @@ function ce(a) {
   }
   return { dataId: s.dataId, shape: c, dtype: s.dtype };
 }
-const Zd = {
+const Jd = {
   kernelName: Qi,
   backendName: "cpu",
   kernelFunc: ce
@@ -323,13 +323,13 @@ function ni(a) {
   for (let H = 0; H < R; H++) {
     const K = H % x, Q = H % g;
     for (let q = 0; q < w; q += M) {
-      const Z = Math.min(q + M, w);
-      for (let J = 0; J < S; J += M) {
-        const ne = Math.min(J + M, S);
+      const J = Math.min(q + M, w);
+      for (let Z = 0; Z < S; Z += M) {
+        const ne = Math.min(Z + M, S);
         for (let re = 0; re < v; re += M) {
           const fe = Math.min(re + M, v);
-          for (let ie = q; ie < Z; ie++)
-            for (let oe = J; oe < ne; oe++) {
+          for (let ie = q; ie < J; ie++)
+            for (let oe = Z; oe < ne; oe++) {
               let de = 0;
               for (let le = re; le < fe; le++) {
                 const Me = (
@@ -349,7 +349,7 @@ function ni(a) {
   }
   return e.disposeIntermediateTensorInfo(k), e.disposeIntermediateTensorInfo($), e.makeTensorInfo(I, W.dtype, W.values);
 }
-const Jd = {
+const Zd = {
   kernelName: lo,
   backendName: "cpu",
   kernelFunc: ni
@@ -609,14 +609,14 @@ function ai(a, t, e, n, s, o) {
             for (; M < 0; )
               M += p;
             const H = Math.min(s.inWidth, f + z), K = B + W * v;
-            let Q = C, q = 0, Z = 0;
+            let Q = C, q = 0, J = 0;
             for (let ne = T; ne < E; ne += l) {
               const re = R + ne * n[1];
               for (let fe = V; fe < L; fe += u) {
                 const ie = re + fe * n[2];
                 for (let oe = M; oe < H; oe += p) {
                   const de = ie + oe * n[3], le = a[de + P];
-                  if (o === "max" && le > Q ? Q = le : o === "avg" && (q += le, Z++), isNaN(Q))
+                  if (o === "max" && le > Q ? Q = le : o === "avg" && (q += le, J++), isNaN(Q))
                     break;
                 }
                 if (isNaN(Q))
@@ -625,8 +625,8 @@ function ai(a, t, e, n, s, o) {
               if (isNaN(Q))
                 break;
             }
-            const J = K + P;
-            b[J] = o === "avg" ? q / Math.max(Z, 1) : Q;
+            const Z = K + P;
+            b[Z] = o === "avg" ? q / Math.max(J, 1) : Q;
           }
         }
       }
@@ -676,7 +676,7 @@ function Rp(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t;
   j(s, "avgPool");
   const { filterSize: o, strides: r, pad: c, dimRoundingMode: i } = n, l = 1;
-  te(Ze(r, l), () => `Error in avgPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
+  te(Je(r, l), () => `Error in avgPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
   const u = Le(s.shape, o, r, l, c, i);
   let p;
   if (u.filterWidth === 1 && u.filterHeight === 1 && nt(u.inShape, u.outShape))
@@ -695,7 +695,7 @@ const Tp = {
 function Ep(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { filterSize: o, strides: r, pad: c, dimRoundingMode: i, dataFormat: l } = n;
   j(s, "avgPool3d");
-  const u = Je(s.shape, o, r, 1, c, i, l), p = e.data.get(s.dataId).values, d = ai(p, s.shape, s.dtype, ee(s.shape), u, "avg");
+  const u = Ze(s.shape, o, r, 1, c, i, l), p = e.data.get(s.dataId).values, d = ai(p, s.shape, s.dtype, ee(s.shape), u, "avg");
   return e.makeTensorInfo(d.shape, "float32", d.values);
 }
 const Dp = {
@@ -706,7 +706,7 @@ const Dp = {
 function Fp(a) {
   const { inputs: t, backend: e, attrs: n } = a, { dy: s, input: o } = t, { filterSize: r, strides: c, pad: i, dimRoundingMode: l } = n;
   j([s, o], "avgPool3DGrad");
-  const u = Je(o.shape, r, c, 1, i, l), p = u.strideDepth, d = u.strideHeight, h = u.strideWidth, f = u.filterDepth, m = u.filterHeight, x = u.filterWidth, g = u.dilationDepth, C = u.dilationHeight, I = u.dilationWidth, b = u.effectiveFilterDepth, y = u.effectiveFilterHeight, k = u.effectiveFilterWidth, $ = b - 1 - u.padInfo.front, v = k - 1 - u.padInfo.left, w = y - 1 - u.padInfo.top, S = pe(o.shape, "float32"), R = 1 / (f * m * x), P = e.bufferSync(s);
+  const u = Ze(o.shape, r, c, 1, i, l), p = u.strideDepth, d = u.strideHeight, h = u.strideWidth, f = u.filterDepth, m = u.filterHeight, x = u.filterWidth, g = u.dilationDepth, C = u.dilationHeight, I = u.dilationWidth, b = u.effectiveFilterDepth, y = u.effectiveFilterHeight, k = u.effectiveFilterWidth, $ = b - 1 - u.padInfo.front, v = k - 1 - u.padInfo.left, w = y - 1 - u.padInfo.top, S = pe(o.shape, "float32"), R = 1 / (f * m * x), P = e.bufferSync(s);
   for (let D = 0; D < u.batchSize; ++D)
     for (let F = 0; F < u.inChannels; ++F)
       for (let T = 0; T < u.inDepth; ++T)
@@ -890,12 +890,12 @@ function ri(a) {
           continue;
         const H = z * k[0], K = O + M * v;
         for (let Q = 0; Q < d.outWidth; ++Q) {
-          const q = B + Q * D, Z = Q * d.strideWidth - g;
-          for (let J = 0; J < f; ++J) {
-            const ne = Z + J * x;
+          const q = B + Q * D, J = Q * d.strideWidth - g;
+          for (let Z = 0; Z < f; ++Z) {
+            const ne = J + Z * x;
             if (ne < 0 || ne >= d.inWidth)
               continue;
-            const re = H + J * k[1], fe = K + ne * w;
+            const re = H + Z * k[1], fe = K + ne * w;
             let ie = re;
             for (let oe = 0; oe < d.inChannels; ++oe) {
               const de = T[fe + oe * S];
@@ -940,12 +940,12 @@ function Qp(a) {
   }
   return e.makeTensorInfo(C.shape, C.dtype, C.values);
 }
-const Zp = {
+const Jp = {
   kernelName: To,
   backendName: "cpu",
   kernelFunc: Qp
 };
-function Jp(a) {
+function Zp(a) {
   const { inputs: t, backend: e, attrs: n } = a, { dy: s, filter: o } = t, { inputShape: r, strides: c, pad: i, dataFormat: l, dimRoundingMode: u } = n;
   j([s, o], "conv2dBackpropInput");
   const p = ee(o.shape), d = ee(s.shape);
@@ -955,13 +955,13 @@ function Jp(a) {
   const A = $ - 1 - f.padInfo.top, N = v - 1 - f.padInfo.left, O = h === "channelsLast", V = m.strides[0], L = O ? m.strides[1] : m.strides[2], B = O ? m.strides[2] : 1, W = O ? 1 : m.strides[1], z = d[0], M = O ? d[1] : d[2], H = O ? d[2] : 1, K = O ? 1 : d[1];
   for (let Q = 0; Q < k; ++Q)
     for (let q = 0; q < w; ++q)
-      for (let Z = 0; Z < S; ++Z) {
-        const J = Z - A, ne = Math.max(0, Math.ceil(J / T)), re = Math.min(D, ($ + J) / T);
+      for (let J = 0; J < S; ++J) {
+        const Z = J - A, ne = Math.max(0, Math.ceil(Z / T)), re = Math.min(D, ($ + Z) / T);
         for (let fe = 0; fe < R; ++fe) {
           const ie = fe - N, oe = Math.max(0, Math.ceil(ie / E)), de = Math.min(F, (v + ie) / E);
           let le = 0;
           for (let me = ne; me < re; ++me) {
-            const He = me * T - J;
+            const He = me * T - Z;
             for (let we = oe; we < de; ++we) {
               const et = we * E - ie, Ee = z * Q + M * me + H * we, Be = I * ($ - 1 - He) + b * (v - 1 - et) + y * q;
               for (let Xe = 0; Xe < P; ++Xe) {
@@ -970,7 +970,7 @@ function Jp(a) {
               }
             }
           }
-          const Me = V * Q + L * Z + B * fe + W * q;
+          const Me = V * Q + L * J + B * fe + W * q;
           x[Me] = le;
         }
       }
@@ -979,7 +979,7 @@ function Jp(a) {
 const eh = {
   kernelName: Eo,
   backendName: "cpu",
-  kernelFunc: Jp
+  kernelFunc: Zp
 };
 function th(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s, filter: o } = t, { strides: r, pad: c, dilations: i } = n;
@@ -1003,11 +1003,11 @@ function th(a) {
             const M = N + W * w[1], H = O + z * v[2];
             for (let K = 0; K < l.outWidth; ++K) {
               const Q = L + K * l.outChannels, q = K * l.strideWidth - C;
-              for (let Z = 0; Z < d; ++Z) {
-                const J = q + Z * m;
-                if (J < 0 || J >= l.inWidth)
+              for (let J = 0; J < d; ++J) {
+                const Z = q + J * m;
+                if (Z < 0 || Z >= l.inWidth)
                   continue;
-                const ne = M + Z * w[2], re = H + J * l.inChannels;
+                const ne = M + J * w[2], re = H + Z * l.inChannels;
                 let fe = ne;
                 for (let ie = 0; ie < l.inChannels; ++ie) {
                   const oe = y[re + ie];
@@ -1038,7 +1038,7 @@ function nh(a) {
     for (let M = 0; M < x; ++M) {
       const H = Math.max(0, Math.ceil((V - M) / h)), K = Math.min(p.outHeight, (p.inHeight + V - M) / h), Q = M * y + z;
       for (let q = 0; q < g; ++q) {
-        const Z = Math.max(0, Math.ceil((O - q) / f)), J = Math.min(p.outWidth, (p.inWidth + O - q) / f), ne = q * k + Q;
+        const J = Math.max(0, Math.ceil((O - q) / f)), Z = Math.min(p.outWidth, (p.inWidth + O - q) / f), ne = q * k + Q;
         for (let re = 0; re < p.inChannels; ++re) {
           const fe = re * $ + ne;
           for (let ie = 0; ie < p.outChannels; ++ie) {
@@ -1049,7 +1049,7 @@ function nh(a) {
                 const we = (L + me * d - N) * T + le, et = me * S + Me;
                 for (let Ee = H; Ee < K; ++Ee) {
                   const Xe = (M + Ee * h - V) * E + we, Ke = Ee * R + et;
-                  for (let je = Z; je < J; ++je) {
+                  for (let je = J; je < Z; ++je) {
                     const ps = (q + je * f - O) * A + Xe, hs = je * P + Ke;
                     oe += D[ps + re] * v[hs + ie];
                   }
@@ -1073,8 +1073,8 @@ function ah(a) {
   const { inputs: t, backend: e, attrs: n } = a, { dy: s, filter: o } = t, { pad: r, strides: c, inputShape: i } = n;
   j([s], "conv3dBackpropInputV2");
   const l = ee(s.shape), u = ee(o.shape), p = xt(i, o.shape, c, 1, r), d = new $e(p.inShape, "float32"), h = d.values, [f, m, x, g] = d.strides, C = e.data.get(s.dataId).values, [I, b, y, k] = l, $ = e.data.get(o.dataId).values, [v, w, S, R] = u, { batchSize: P, filterDepth: D, filterHeight: F, filterWidth: T, inChannels: E, inDepth: A, inHeight: N, inWidth: O, outChannels: V, outDepth: L, outHeight: B, outWidth: W, strideDepth: z, strideHeight: M, strideWidth: H } = p, K = D - 1 - p.padInfo.front, Q = F - 1 - p.padInfo.top, q = T - 1 - p.padInfo.left;
-  for (let Z = 0; Z < P; ++Z)
-    for (let J = 0; J < E; ++J)
+  for (let J = 0; J < P; ++J)
+    for (let Z = 0; Z < E; ++Z)
       for (let ne = 0; ne < A; ++ne) {
         const re = ne - K, fe = Math.max(0, Math.ceil(re / z)), ie = Math.min(L, (D + re) / z);
         for (let oe = 0; oe < N; ++oe) {
@@ -1087,7 +1087,7 @@ function ah(a) {
               for (let Ke = le; Ke < Me; ++Ke) {
                 const je = Ke * M - de;
                 for (let kt = we; kt < et; ++kt) {
-                  const ps = kt * H - He, hs = I * Z + b * Be + y * Ke + k * kt, Bi = v * (D - 1 - Xe) + w * (F - 1 - je) + S * (T - 1 - ps) + R * J;
+                  const ps = kt * H - He, hs = I * J + b * Be + y * Ke + k * kt, Bi = v * (D - 1 - Xe) + w * (F - 1 - je) + S * (T - 1 - ps) + R * Z;
                   for (let Bt = 0; Bt < V; ++Bt) {
                     const Gi = C[hs + Bt], zi = $[Bi + Bt];
                     Ee += Gi * zi;
@@ -1095,7 +1095,7 @@ function ah(a) {
                 }
               }
             }
-            h[f * Z + m * ne + x * oe + g * me + J] = Ee;
+            h[f * J + m * ne + x * oe + g * me + Z] = Ee;
           }
         }
       }
@@ -1151,9 +1151,9 @@ function dh(a) {
             K = H + z * y[2] + N * y[1] + D * y[0];
             const q = b[K];
             K = H + W * y[2] + O * y[1] + D * y[0];
-            const Z = b[K];
+            const J = b[K];
             K = H + z * y[2] + O * y[1] + D * y[0];
-            const J = b[K], ne = Q + (q - Q) * M, re = Z + (J - Z) * M;
+            const Z = b[K], ne = Q + (q - Q) * M, re = J + (Z - J) * M;
             K = H + L * k[2] + E * k[1] + $ * k[0], g.values[K] = ne + (re - ne) * V;
           }
         }
@@ -1191,7 +1191,7 @@ function hh(a) {
   const u = xe(1, s.shape.length)[0];
   if (u !== l.shape.length - 1)
     throw new Error(`backend.cumprod in CPU expects an inner-most axis=${l.shape.length - 1} but got axis=${u}`);
-  const p = Qe(l.dtype, "int32"), d = Ji(G(l.shape), p), h = e.data.get(l.dataId).values, f = l.shape[l.shape.length - 1], m = c ? (g, C) => g + f - C - 1 : (g, C) => g + C;
+  const p = Qe(l.dtype, "int32"), d = Zi(G(l.shape), p), h = e.data.get(l.dataId).values, f = l.shape[l.shape.length - 1], m = c ? (g, C) => g + f - C - 1 : (g, C) => g + C;
   for (let g = 0; g < h.length; g += f)
     for (let C = 0; C < f; C++) {
       const I = m(g, C);
@@ -1290,7 +1290,7 @@ function ii(a) {
   j([s, o], "depthwiseConv2DNative");
   const u = ee(s.shape), p = ee(o.shape);
   let d = i;
-  d == null && (d = [1, 1]), te(Ze(r, d), () => `Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${r} and dilations '${d}'`);
+  d == null && (d = [1, 1]), te(Je(r, d), () => `Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${r} and dilations '${d}'`);
   const h = Se(
     s.shape,
     o.shape,
@@ -1319,9 +1319,9 @@ function ii(a) {
             const M = N + W * p[1], H = O + z * h.inChannels;
             let K = L, Q = M;
             for (let q = 0; q < h.inChannels; ++q) {
-              const Z = $[H + q];
-              for (let J = 0; J < y; ++J)
-                w[K + J] += Z * v[Q + J];
+              const J = $[H + q];
+              for (let Z = 0; Z < y; ++Z)
+                w[K + Z] += J * v[Q + Z];
               K += y, Q += y;
             }
           }
@@ -1391,11 +1391,11 @@ function vh(a) {
   for (let M = 0; M < R; ++M)
     for (let H = 0; H < F; ++H)
       for (let K = 0; K < T; ++K) {
-        const Q = K - B, q = Math.max(0, Math.ceil(Q / V)), Z = Math.min(N, (P + Q) / V);
-        for (let J = 0; J < E; ++J) {
-          const ne = J - W, re = Math.max(0, Math.ceil(ne / L)), fe = Math.min(O, (D + ne) / L);
+        const Q = K - B, q = Math.max(0, Math.ceil(Q / V)), J = Math.min(N, (P + Q) / V);
+        for (let Z = 0; Z < E; ++Z) {
+          const ne = Z - W, re = Math.max(0, Math.ceil(ne / L)), fe = Math.min(O, (D + ne) / L);
           let ie = 0;
-          for (let oe = q; oe < Z; ++oe) {
+          for (let oe = q; oe < J; ++oe) {
             const de = oe * V - Q;
             for (let le = re; le < fe; ++le) {
               const Me = le * L - ne, me = b * M + y * oe + k * le, He = v * (P - 1 - de) + w * (D - 1 - Me) + S * H;
@@ -1405,7 +1405,7 @@ function vh(a) {
               }
             }
           }
-          m[x * M + g * K + C * J + H] = ie;
+          m[x * M + g * K + C * Z + H] = ie;
         }
       }
   return e.makeTensorInfo(f.shape, f.dtype, f.values);
@@ -1639,7 +1639,7 @@ const Vh = Sr, Wh = vr, Mh = $r, Bh = kr, Gh = yr, zh = Ir, Uh = ae(Ms, (a) => {
   backendName: "cpu",
   kernelFunc: Uh
 };
-function Zt(a) {
+function Jt(a) {
   const { inputs: t, backend: e, attrs: n } = a, { input: s } = t, { dim: o } = n, r = s.shape.length, c = s.shape.slice();
   let i = o;
   return o < 0 && (te(-(r + 1) <= o, () => `Axis must be in the interval [${-(r + 1)}, ${r}]`), i = r + o + 1), c.splice(i, 0, 1), ce({ inputs: { x: s }, backend: e, attrs: { shape: c } });
@@ -1647,7 +1647,7 @@ function Zt(a) {
 const Xh = {
   kernelName: Ko,
   backendName: "cpu",
-  kernelFunc: Zt
+  kernelFunc: Jt
 };
 const Kh = ot((a, t) => a / t), gn = gt(jo, Kh), Is = {
   kernelName: jo,
@@ -1704,16 +1704,16 @@ function ys(a, t, e, n, s) {
   }), H = ur({
     inputs: { a: R, b: z },
     backend: s
-  }), K = Nt({ inputs: { input: M }, backend: s }), Q = Nt({ inputs: { input: H }, backend: s }), q = ht({ inputs: { input: M }, backend: s }), Z = ht({ inputs: { input: H }, backend: s }), J = ft({
+  }), K = Nt({ inputs: { input: M }, backend: s }), Q = Nt({ inputs: { input: H }, backend: s }), q = ht({ inputs: { input: M }, backend: s }), J = ht({ inputs: { input: H }, backend: s }), Z = ft({
     inputs: [K, Q],
     backend: s,
     attrs: { axis: 0 }
   }), ne = ft({
-    inputs: [q, Z],
+    inputs: [q, J],
     backend: s,
     attrs: { axis: 0 }
-  }), re = s.data.get(J.dataId).values, fe = s.data.get(ne.dataId).values;
-  return s.disposeIntermediateTensorInfo(p), s.disposeIntermediateTensorInfo(d), s.disposeIntermediateTensorInfo(h), s.disposeIntermediateTensorInfo(C), s.disposeIntermediateTensorInfo(I), s.disposeIntermediateTensorInfo(b), s.disposeIntermediateTensorInfo(w), s.disposeIntermediateTensorInfo(S), s.disposeIntermediateTensorInfo(R), s.disposeIntermediateTensorInfo(E), s.disposeIntermediateTensorInfo(A), s.disposeIntermediateTensorInfo(N), s.disposeIntermediateTensorInfo(L), s.disposeIntermediateTensorInfo(B), s.disposeIntermediateTensorInfo(W), s.disposeIntermediateTensorInfo(z), s.disposeIntermediateTensorInfo(M), s.disposeIntermediateTensorInfo(H), s.disposeIntermediateTensorInfo(K), s.disposeIntermediateTensorInfo(q), s.disposeIntermediateTensorInfo(Q), s.disposeIntermediateTensorInfo(Z), s.disposeIntermediateTensorInfo(J), s.disposeIntermediateTensorInfo(ne), { real: re, imag: fe };
+  }), re = s.data.get(Z.dataId).values, fe = s.data.get(ne.dataId).values;
+  return s.disposeIntermediateTensorInfo(p), s.disposeIntermediateTensorInfo(d), s.disposeIntermediateTensorInfo(h), s.disposeIntermediateTensorInfo(C), s.disposeIntermediateTensorInfo(I), s.disposeIntermediateTensorInfo(b), s.disposeIntermediateTensorInfo(w), s.disposeIntermediateTensorInfo(S), s.disposeIntermediateTensorInfo(R), s.disposeIntermediateTensorInfo(E), s.disposeIntermediateTensorInfo(A), s.disposeIntermediateTensorInfo(N), s.disposeIntermediateTensorInfo(L), s.disposeIntermediateTensorInfo(B), s.disposeIntermediateTensorInfo(W), s.disposeIntermediateTensorInfo(z), s.disposeIntermediateTensorInfo(M), s.disposeIntermediateTensorInfo(H), s.disposeIntermediateTensorInfo(K), s.disposeIntermediateTensorInfo(q), s.disposeIntermediateTensorInfo(Q), s.disposeIntermediateTensorInfo(J), s.disposeIntermediateTensorInfo(Z), s.disposeIntermediateTensorInfo(ne), { real: re, imag: fe };
 }
 function Yh(a, t, e) {
   const n = new Float32Array(t * 2);
@@ -1735,7 +1735,7 @@ function Qh(a) {
   }), i = ci(c, !1, e), l = ce({ inputs: { x: i }, backend: e, attrs: { shape: n.shape } });
   return e.disposeIntermediateTensorInfo(c), e.disposeIntermediateTensorInfo(i), l;
 }
-const Zh = {
+const Jh = {
   kernelName: qo,
   backendName: "cpu",
   kernelFunc: Qh
@@ -1744,7 +1744,7 @@ function xn(a) {
   const { backend: t, attrs: e } = a, { shape: n, value: s, dtype: o } = e, r = o || Qo(s), c = ns(r, G(n));
   return ef(c, s, r), t.makeTensorInfo(n, r, c);
 }
-const Jh = {
+const Zh = {
   kernelName: Yo,
   backendName: "cpu",
   kernelFunc: xn
@@ -1753,7 +1753,7 @@ function ef(a, t, e) {
   a.fill(t);
 }
 const tf = {
-  kernelName: Zo,
+  kernelName: Jo,
   backendName: "cpu",
   kernelFunc: ({ inputs: a, attrs: t, backend: e }) => {
     const { image: n } = a, s = e, o = _e(n.dtype, G(n.shape)), [r, c, i, l] = n.shape, u = s.data.get(n.dataId).values;
@@ -1810,7 +1810,7 @@ function sf(a) {
   return m;
 }
 const nf = {
-  kernelName: Jo,
+  kernelName: Zo,
   backendName: "cpu",
   kernelFunc: sf
 };
@@ -2018,7 +2018,7 @@ function _f(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t;
   j(s, "maxPool");
   const { filterSize: o, strides: r, pad: c, dimRoundingMode: i } = n, l = 1;
-  te(Ze(r, l), () => `Error in maxPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
+  te(Je(r, l), () => `Error in maxPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
   const u = Le(s.shape, o, r, l, c, i);
   let p;
   if (u.filterWidth === 1 && u.filterHeight === 1 && nt(u.inShape, u.outShape))
@@ -2037,7 +2037,7 @@ const Lf = {
 function Vf(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { filterSize: o, strides: r, pad: c, dimRoundingMode: i, dataFormat: l } = n;
   j(s, "maxPool3d");
-  const u = Je(s.shape, o, r, 1, c, i, l), p = e.data.get(s.dataId).values, d = ai(p, s.shape, s.dtype, ee(s.shape), u, "max");
+  const u = Ze(s.shape, o, r, 1, c, i, l), p = e.data.get(s.dataId).values, d = ai(p, s.shape, s.dtype, ee(s.shape), u, "max");
   return e.makeTensorInfo(d.shape, "float32", d.values);
 }
 const Wf = {
@@ -2048,7 +2048,7 @@ const Wf = {
 function Mf(a) {
   const { inputs: t, backend: e, attrs: n } = a, { dy: s, input: o } = t, { filterSize: r, strides: c, pad: i, dimRoundingMode: l } = n;
   j([s, o], "maxPool3DGrad");
-  const u = Je(o.shape, r, c, 1, i, l), p = e.bufferSync(o), d = Np(p, u), h = u.strideDepth, f = u.strideHeight, m = u.strideWidth, x = u.dilationDepth, g = u.dilationHeight, C = u.dilationWidth, I = u.effectiveFilterDepth, b = u.effectiveFilterHeight, y = u.effectiveFilterWidth, k = I - 1 - u.padInfo.front, $ = y - 1 - u.padInfo.left, v = b - 1 - u.padInfo.top, w = pe(o.shape, "float32"), S = e.bufferSync(s);
+  const u = Ze(o.shape, r, c, 1, i, l), p = e.bufferSync(o), d = Np(p, u), h = u.strideDepth, f = u.strideHeight, m = u.strideWidth, x = u.dilationDepth, g = u.dilationHeight, C = u.dilationWidth, I = u.effectiveFilterDepth, b = u.effectiveFilterHeight, y = u.effectiveFilterWidth, k = I - 1 - u.padInfo.front, $ = y - 1 - u.padInfo.left, v = b - 1 - u.padInfo.top, w = pe(o.shape, "float32"), S = e.bufferSync(s);
   for (let R = 0; R < u.batchSize; ++R)
     for (let P = 0; P < u.inChannels; ++P)
       for (let D = 0; D < u.inDepth; ++D)
@@ -2201,13 +2201,13 @@ const Qf = {
   backendName: "cpu",
   kernelFunc: Yf
 };
-const Zf = ot(((a, t) => {
+const Jf = ot(((a, t) => {
   const e = a % t;
   return a < 0 && t < 0 || a >= 0 && t >= 0 ? e : (e + t) % t;
-})), Jf = gt(js, Zf), em = {
+})), Zf = gt(js, Jf), em = {
   kernelName: js,
   backendName: "cpu",
-  kernelFunc: Jf
+  kernelFunc: Zf
 };
 function ui(a) {
   const { inputs: t, backend: e, attrs: n } = a, { logits: s } = t, { dim: o } = n, r = s.shape.length;
@@ -2235,7 +2235,7 @@ function sm(a) {
     x[0] = p[m];
     for (let I = 1; I < x.length; ++I)
       x[I] = x[I - 1] + p[m + I];
-    const g = Jl.alea(r.toString()), C = f * o;
+    const g = Zl.alea(r.toString()), C = f * o;
     for (let I = 0; I < o; ++I) {
       const b = g();
       h[C + I] = x.length;
@@ -2310,12 +2310,12 @@ const fm = {
   backendName: "cpu",
   kernelFunc: hm
 };
-function Jt(a) {
+function Zt(a) {
   const { inputs: t, backend: e } = a, { x: n } = t;
   if (n.dtype === "string")
     throw new Error("zerosLike is not supported for string tensors");
   if (n.dtype === "complex64") {
-    const s = Nt({ inputs: { input: n }, backend: e }), o = Jt({ inputs: { x: s }, backend: e }), r = ht({ inputs: { input: n }, backend: e }), c = Jt({ inputs: { x: r }, backend: e }), i = Ae({ inputs: { real: o, imag: c }, backend: e });
+    const s = Nt({ inputs: { input: n }, backend: e }), o = Zt({ inputs: { x: s }, backend: e }), r = ht({ inputs: { input: n }, backend: e }), c = Zt({ inputs: { x: r }, backend: e }), i = Ae({ inputs: { real: o, imag: c }, backend: e });
     return e.disposeIntermediateTensorInfo(s), e.disposeIntermediateTensorInfo(o), e.disposeIntermediateTensorInfo(r), e.disposeIntermediateTensorInfo(c), i;
   } else
     return xn({ backend: e, attrs: { shape: n.shape, value: 0, dtype: n.dtype } });
@@ -2323,14 +2323,14 @@ function Jt(a) {
 const mm = {
   kernelName: ya,
   backendName: "cpu",
-  kernelFunc: Jt
+  kernelFunc: Zt
 };
 function di(a) {
   const { inputs: t, backend: e } = a, { x: n } = t;
   if (n.dtype === "string")
     throw new Error("onesLike is not supported for string tensors");
   if (n.dtype === "complex64") {
-    const s = Nt({ inputs: { input: n }, backend: e }), o = di({ inputs: { x: s }, backend: e }), r = ht({ inputs: { input: n }, backend: e }), c = Jt({ inputs: { x: r }, backend: e }), i = Ae({ inputs: { real: o, imag: c }, backend: e });
+    const s = Nt({ inputs: { input: n }, backend: e }), o = di({ inputs: { x: s }, backend: e }), r = ht({ inputs: { input: n }, backend: e }), c = Zt({ inputs: { x: r }, backend: e }), i = Ae({ inputs: { real: o, imag: c }, backend: e });
     return e.disposeIntermediateTensorInfo(s), e.disposeIntermediateTensorInfo(o), e.disposeIntermediateTensorInfo(r), e.disposeIntermediateTensorInfo(c), i;
   } else
     return xn({ backend: e, attrs: { shape: n.shape, value: 1, dtype: n.dtype } });
@@ -2343,13 +2343,13 @@ const gm = {
 function pi(a) {
   const { inputs: t, backend: e, attrs: n } = a, { axis: s } = n;
   if (t.length === 1)
-    return Zt({ inputs: { input: t[0] }, backend: e, attrs: { dim: s } });
+    return Jt({ inputs: { input: t[0] }, backend: e, attrs: { dim: s } });
   const o = t[0].shape, r = t[0].dtype;
   t.forEach((u) => {
     va(o, u.shape, "All tensors passed to stack must have matching shapes"), te(r === u.dtype, () => "All tensors passed to stack must have matching dtypes");
   });
   const c = [], i = t.map((u) => {
-    const p = Zt({ inputs: { input: u }, backend: e, attrs: { dim: s } });
+    const p = Jt({ inputs: { input: u }, backend: e, attrs: { dim: s } });
     return c.push(p), p;
   }), l = ft({ inputs: i, backend: e, attrs: { axis: s } });
   return c.forEach((u) => e.disposeIntermediateTensorInfo(u)), l;
@@ -2384,7 +2384,7 @@ const bm = ot((a, t) => Math.pow(a, t)), Im = gt(qs, bm), ym = {
   kernelFunc: Im
 };
 function km(a) {
-  const { inputs: t, backend: e, attrs: n } = a, { paramsNestedSplits: s, paramsDenseValues: o, indices: r } = t, { outputRaggedRank: c } = n, i = s.map((g) => e.data.get(g.dataId).values), l = s.map((g) => g.shape), u = e.data.get(o.dataId).values, p = e.data.get(r.dataId).values, [d, h, f] = Zc(i, l, u, o.shape, o.dtype, p, r.shape), m = d.map((g) => e.makeTensorInfo([g.length], "int32", g)), x = e.makeTensorInfo(f, o.dtype, h);
+  const { inputs: t, backend: e, attrs: n } = a, { paramsNestedSplits: s, paramsDenseValues: o, indices: r } = t, { outputRaggedRank: c } = n, i = s.map((g) => e.data.get(g.dataId).values), l = s.map((g) => g.shape), u = e.data.get(o.dataId).values, p = e.data.get(r.dataId).values, [d, h, f] = Jc(i, l, u, o.shape, o.dtype, p, r.shape), m = d.map((g) => e.makeTensorInfo([g.length], "int32", g)), x = e.makeTensorInfo(f, o.dtype, h);
   return m.concat([x]);
 }
 const $m = {
@@ -2393,7 +2393,7 @@ const $m = {
   kernelFunc: km
 };
 function vm(a) {
-  const { inputs: t, backend: e } = a, { starts: n, limits: s, deltas: o } = t, r = e.data.get(n.dataId).values, c = e.data.get(s.dataId).values, i = e.data.get(o.dataId).values, [l, u] = Jc(r, n.shape, n.dtype, c, s.shape, i, o.shape), p = e.makeTensorInfo([l.length], "int32", l), d = e.makeTensorInfo([u.length], n.dtype, u);
+  const { inputs: t, backend: e } = a, { starts: n, limits: s, deltas: o } = t, r = e.data.get(n.dataId).values, c = e.data.get(s.dataId).values, i = e.data.get(o.dataId).values, [l, u] = Zc(r, n.shape, n.dtype, c, s.shape, i, o.shape), p = e.makeTensorInfo([l.length], "int32", l), d = e.makeTensorInfo([u.length], n.dtype, u);
   return [p, d];
 }
 const Sm = {
@@ -2446,8 +2446,8 @@ function Fm(a) {
         r ? T = y * (F + 0.5) - 0.5 : T = y * F;
         const E = Math.max(0, Math.floor(T)), A = T - E, N = Math.min(h - 1, Math.ceil(T)), O = P + E * i[2], V = D + E * i[2], L = P + N * i[2], B = D + N * i[2];
         for (let W = 0; W < f; W++) {
-          const z = m[O + W], M = m[V + W], H = m[L + W], K = m[B + W], Q = z + (H - z) * A, q = M + (K - M) * A, Z = Q + (q - Q) * S;
-          x[I++] = Z;
+          const z = m[O + W], M = m[V + W], H = m[L + W], K = m[B + W], Q = z + (H - z) * A, q = M + (K - M) * A, J = Q + (q - Q) * S;
+          x[I++] = J;
         }
       }
     }
@@ -2476,8 +2476,8 @@ function Om(a) {
       for (let T = 0; T < h; T++) {
         const E = T * C, A = Math.floor(E), N = Math.min(Math.ceil(E), u - 1), O = E - A, V = 1 - O, L = R + A * c[2], B = R + N * c[2], W = P + A * c[2], z = P + N * c[2], M = F * V, H = F * O, K = D * V, Q = D * O;
         for (let q = 0; q < p; q++) {
-          const Z = I[b++];
-          f[L + q] += Z * M, f[B + q] += Z * H, f[W + q] += Z * K, f[z + q] += Z * Q;
+          const J = I[b++];
+          f[L + q] += J * M, f[B + q] += J * H, f[W + q] += J * K, f[z + q] += J * Q;
         }
       }
     }
@@ -2554,8 +2554,8 @@ function Vm(a) {
                 const K = H + N;
                 if (K < 0 || K >= f)
                   continue;
-                const Q = W + K * i[2], q = K * b, Z = Math.min(p - 1, r ? Math.round(q) : Math.floor(q));
-                T === Z && (V += x[Q + O]);
+                const Q = W + K * i[2], q = K * b, J = Math.min(p - 1, r ? Math.round(q) : Math.floor(q));
+                T === J && (V += x[Q + O]);
               }
           }
           m[E + O] = V;
@@ -2664,7 +2664,7 @@ const Qm = {
   backendName: "cpu",
   kernelFunc: Ym
 };
-function Zm(a) {
+function Jm(a) {
   const { inputs: t, backend: e } = a, { condition: n, t: s, e: o } = t;
   j([n, s, o], "select");
   const r = n.shape.length, c = e.data.get(n.dataId).values, i = e.data.get(s.dataId).values, l = e.data.get(o.dataId).values, u = Qe(s.dtype, o.dtype), p = Ye(G(s.shape), u);
@@ -2675,18 +2675,18 @@ function Zm(a) {
       c[f] === 1 ? p[d++] = i[f] : p[d++] = l[f];
   return e.makeTensorInfo(s.shape, u, p);
 }
-const Jm = {
+const Zm = {
   kernelName: Va,
   backendName: "cpu",
-  kernelFunc: Zm
+  kernelFunc: Jm
 };
-const eg = _r, tg = Ar, sg = ae(Zs, (a) => a >= 0 ? tg * a : eg * (Math.exp(a) - 1)), ng = {
-  kernelName: Zs,
+const eg = _r, tg = Ar, sg = ae(Js, (a) => a >= 0 ? tg * a : eg * (Math.exp(a) - 1)), ng = {
+  kernelName: Js,
   backendName: "cpu",
   kernelFunc: sg
 };
-const og = ae(Js, (a) => a < 0 ? -1 : a > 0 ? 1 : 0), ag = {
-  kernelName: Js,
+const og = ae(Zs, (a) => a < 0 ? -1 : a > 0 ? 1 : 0), ag = {
+  kernelName: Zs,
   backendName: "cpu",
   kernelFunc: og
 };
@@ -2980,7 +2980,7 @@ function Ug(a) {
   return e.makeTensorInfo(r.shape, r.dtype, r.values);
 }
 const Hg = {
-  kernelName: Za,
+  kernelName: Ja,
   backendName: "cpu",
   kernelFunc: Ug
 };
@@ -2994,7 +2994,7 @@ function Xg(a) {
   ];
 }
 const Kg = {
-  kernelName: Ja,
+  kernelName: Za,
   backendName: "cpu",
   kernelFunc: Xg
 };
@@ -3041,9 +3041,9 @@ function _n(a, t, e) {
     case "wrap":
       return Qg(a, t);
     case "nearest":
-      return Jg(a, t);
+      return Zg(a, t);
     default:
-      return Zg(a);
+      return Jg(a);
   }
 }
 function Yg(a, t) {
@@ -3082,10 +3082,10 @@ function Qg(a, t) {
     }
   return rn(0, e, t - 1);
 }
-function Zg(a, t) {
+function Jg(a, t) {
   return a;
 }
-function Jg(a, t) {
+function Zg(a, t) {
   return rn(0, a, t - 1);
 }
 function $t(a, t, e, n, s, o, r, c, i, l, u) {
@@ -3143,7 +3143,7 @@ function rx(a) {
   const c = s.shape.length, i = o.shape.length, l = [], u = [], p = c - i;
   let d = o;
   for (let f = 0; f < p; ++f) {
-    const m = Zt({ inputs: { input: d }, backend: e, attrs: { dim: f + 1 } });
+    const m = Jt({ inputs: { input: d }, backend: e, attrs: { dim: f + 1 } });
     d = m, u.push(m);
   }
   for (let f = 0; f < r; ++f) {
@@ -3178,7 +3178,7 @@ const cx = [
   Dp,
   Pp,
   Ap,
-  Jd,
+  Zd,
   Lp,
   Wp,
   Bp,
@@ -3191,7 +3191,7 @@ const cx = [
   Kp,
   qp,
   Yp,
-  Zp,
+  Jp,
   eh,
   sh,
   oh,
@@ -3219,8 +3219,8 @@ const cx = [
   bl,
   Xh,
   Il,
-  Zh,
   Jh,
+  Zh,
   tf,
   yl,
   kl,
@@ -3282,7 +3282,7 @@ const cx = [
   Dm,
   Yd,
   Qd,
-  Zd,
+  Jd,
   Pm,
   Am,
   Lm,
@@ -3293,7 +3293,7 @@ const cx = [
   _l,
   Xm,
   Qm,
-  Jm,
+  Zm,
   ng,
   Ll,
   ag,
@@ -4825,15 +4825,15 @@ const Qx = {
   backendName: "webgl",
   kernelFunc: Yx
 };
-const Zx = We + `
+const Jx = We + `
   if (abs(x) > 1.) {
     return NAN;
   }
   return acos(x);
-`, Jx = se({ opSnippet: Zx }), eC = {
+`, Zx = se({ opSnippet: Jx }), eC = {
   kernelName: Ts,
   backendName: "webgl",
-  kernelFunc: Jx
+  kernelFunc: Zx
 };
 const tC = We + `
   if (x < 1.0) return NAN;
@@ -5507,7 +5507,7 @@ function AC(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t;
   At(s, "avgPool");
   const { filterSize: o, strides: r, pad: c, dimRoundingMode: i } = n, l = 1;
-  te(Ze(r, l), () => `Error in avgPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
+  te(Je(r, l), () => `Error in avgPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
   const u = Le(s.shape, o, r, l, c, i);
   if (u.filterWidth === 1 && u.filterHeight === 1 && nt(u.inShape, u.outShape))
     return Fe({ inputs: { x: s }, backend: e });
@@ -5520,7 +5520,7 @@ const _C = {
   kernelFunc: AC
 };
 function LC(a) {
-  const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { filterSize: o, strides: r, pad: c, dimRoundingMode: i, dataFormat: l } = n, u = [1, 1, 1], p = Je(s.shape, o, r, u, c, i, l), d = new bn(p, "avg", !1);
+  const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { filterSize: o, strides: r, pad: c, dimRoundingMode: i, dataFormat: l } = n, u = [1, 1, 1], p = Ze(s.shape, o, r, u, c, i, l), d = new bn(p, "avg", !1);
   return e.runWebGLProgram(d, [s], "float32");
 }
 const VC = {
@@ -5641,7 +5641,7 @@ class MC {
   }
 }
 function BC(a) {
-  const { inputs: t, backend: e, attrs: n } = a, { dy: s, input: o } = t, r = o, { filterSize: c, strides: i, pad: l, dimRoundingMode: u } = n, p = [1, 1, 1], d = Je(r.shape, c, i, p, l, u), h = new MC(d);
+  const { inputs: t, backend: e, attrs: n } = a, { dy: s, input: o } = t, r = o, { filterSize: c, strides: i, pad: l, dimRoundingMode: u } = n, p = [1, 1, 1], d = Ze(r.shape, c, i, p, l, u), h = new MC(d);
   return e.runWebGLProgram(h, [s], r.dtype);
 }
 const GC = {
@@ -5732,7 +5732,7 @@ class QC {
     this.variableNames = ["source"], this.outputShape = t, this.rank = t.length;
     const e = he(this.rank);
     this.customUniforms = [{ name: "start", arrayIndex: this.rank, type: "int" }];
-    const n = ZC(this.rank);
+    const n = JC(this.rank);
     let s;
     const o = t.map((r, c) => `sourceLoc.${ks[c]} = start[${c}] + coords.${ks[c]};`);
     s = `
@@ -5749,14 +5749,14 @@ class QC {
   }
 }
 const ks = ["x", "y", "z", "w", "u", "v"];
-function ZC(a) {
+function JC(a) {
   if (a === 1)
     return "sourceLoc";
   if (a <= 6)
     return ks.slice(0, a).map((t) => "sourceLoc." + t).join(",");
   throw Error(`Slicing for rank ${a} is not yet supported`);
 }
-class JC {
+class ZC {
   constructor(t) {
     this.variableNames = ["source"], this.packedInputs = !0, this.packedOutput = !0, this.outputShape = t, this.rank = t.length, this.customUniforms = [{ name: "start", arrayIndex: this.rank, type: "int" }];
     const e = he(this.rank), n = ye("coords", this.rank), s = ye("sourceLoc", this.rank), o = this.rank === 1 ? "sourceLoc" : `vec2(${s.slice(-2).join()})`, r = `getChannel(getSource(${s.join()}), ${o})`, c = `
@@ -5814,7 +5814,7 @@ function yt(a) {
   }
   const { isPacked: l } = e.texData.get(s.dataId), u = su(s.shape, c, i);
   if (l || !u) {
-    const p = _().getBool("WEBGL_PACK_ARRAY_OPERATIONS") ? new JC(i) : new QC(i), d = [c];
+    const p = _().getBool("WEBGL_PACK_ARRAY_OPERATIONS") ? new ZC(i) : new QC(i), d = [c];
     return e.runWebGLProgram(p, [s], s.dtype, d);
   }
   return e.uploadToGPU(s.dataId), eb(s, c, i, e);
@@ -7233,15 +7233,15 @@ const Yb = {
 };
 const Qb = It + `
   return cos(x);
-`, Zb = `
+`, Jb = `
   vec4 result = cos(x);
   bvec4 isNaN = isnan(x);
   ${it}
   return result;
-`, Jb = se({ opSnippet: Qb, packedOpSnippet: Zb }), e1 = {
+`, Zb = se({ opSnippet: Qb, packedOpSnippet: Jb }), e1 = {
   kernelName: Ls,
   backendName: "webgl",
-  kernelFunc: Jb
+  kernelFunc: Zb
 };
 const t1 = `
   float e2x = exp(-x);
@@ -7791,7 +7791,7 @@ class Di {
 function g1(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s, filter: o } = t, { strides: r, pad: c, dilations: i, dimRoundingMode: l } = n;
   let u = i;
-  u == null && (u = [1, 1]), te(Ze(r, u), () => `Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${r} and dilations '${u}'`);
+  u == null && (u = [1, 1]), te(Je(r, u), () => `Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${r} and dilations '${u}'`);
   const p = Se(
     s.shape,
     o.shape,
@@ -8147,12 +8147,12 @@ const Y1 = {
   backendName: "webgl",
   kernelFunc: vs
 };
-const Qn = "return exp(x) - 1.0;", Q1 = se({ opSnippet: Qn, packedOpSnippet: Qn, cpuKernelImpl: zu }), Z1 = {
+const Qn = "return exp(x) - 1.0;", Q1 = se({ opSnippet: Qn, packedOpSnippet: Qn, cpuKernelImpl: zu }), J1 = {
   kernelName: Ic,
   backendName: "webgl",
   kernelFunc: Q1
 };
-class Zn {
+class Jn {
   constructor(t, e, n) {
     this.variableNames = ["real", "imag"];
     const s = e[1];
@@ -8202,7 +8202,7 @@ class Zn {
   }
 }
 function Pi(a, t, e) {
-  const n = e.texData.get(a.dataId), s = G(a.shape), o = a.shape[a.shape.length - 1], r = s / o, c = U({ inputs: { x: a }, backend: e, attrs: { shape: [r, o] } }), i = c.shape, l = new Zn("real", i, t), u = new Zn("imag", i, t), p = [
+  const n = e.texData.get(a.dataId), s = G(a.shape), o = a.shape[a.shape.length - 1], r = s / o, c = U({ inputs: { x: a }, backend: e, attrs: { shape: [r, o] } }), i = c.shape, l = new Jn("real", i, t), u = new Jn("imag", i, t), p = [
     {
       dataId: n.complexTensorInfos.real.dataId,
       dtype: n.complexTensorInfos.real.dtype,
@@ -8218,14 +8218,14 @@ function Pi(a, t, e) {
   const m = U({ inputs: { x: f }, backend: e, attrs: { shape: a.shape } });
   return e.disposeIntermediateTensorInfo(c), e.disposeIntermediateTensorInfo(f), m;
 }
-function J1(a) {
+function Z1(a) {
   const { inputs: t, backend: e } = a, { input: n } = t;
   return Pi(n, !1, e);
 }
 const e0 = {
   kernelName: qo,
   backendName: "webgl",
-  kernelFunc: J1
+  kernelFunc: Z1
 };
 class t0 {
   constructor(t, e) {
@@ -8275,14 +8275,14 @@ class n0 {
   }
 }
 const o0 = {
-  kernelName: Zo,
+  kernelName: Jo,
   backendName: "webgl",
   kernelFunc: ({ inputs: a, backend: t }) => {
     const { image: e } = a, n = t, s = new n0(e.shape);
     return n.runWebGLProgram(s, [e], e.dtype);
   }
 };
-const Jn = "return floor(x);", a0 = se({ opSnippet: Jn, packedOpSnippet: Jn, cpuKernelImpl: Uu }), r0 = {
+const Zn = "return floor(x);", a0 = se({ opSnippet: Zn, packedOpSnippet: Zn, cpuKernelImpl: Uu }), r0 = {
   kernelName: yc,
   backendName: "webgl",
   kernelFunc: a0
@@ -8474,14 +8474,14 @@ function m0(a) {
   return C.push(g), C.forEach((v) => e.disposeIntermediateTensorInfo(v)), $;
 }
 const g0 = {
-  kernelName: Jo,
+  kernelName: Zo,
   backendName: "webgl",
   kernelFunc: m0
 };
 function x0(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s, filter: o, bias: r, preluActivationWeights: c } = t, { strides: i, pad: l, dilations: u, dimRoundingMode: p, activation: d, leakyreluAlpha: h } = n, f = [];
   let m = u;
-  m == null && (m = [1, 1]), te(Ze(i, m), () => `Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${i} and dilations '${m}'`);
+  m == null && (m = [1, 1]), te(Je(i, m), () => `Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${i} and dilations '${m}'`);
   const x = Se(
     s.shape,
     o.shape,
@@ -8692,14 +8692,14 @@ const j0 = "return float(a <= b);", q0 = `
   backendName: "webgl",
   kernelFunc: Y0
 };
-function Z0(a) {
+function J0(a) {
   const { backend: t, attrs: e } = a, { start: n, stop: s, num: o } = e, r = Qu(n, s, o);
   return t.makeTensorInfo([r.length], "float32", r);
 }
-const J0 = {
+const Z0 = {
   kernelName: oa,
   backendName: "webgl",
-  kernelFunc: Z0
+  kernelFunc: J0
 };
 const eI = It + `
   return x < 0.0 ? 0./0. : log(x);
@@ -8711,7 +8711,7 @@ const eI = It + `
   result.b = isNaN.b ? x.b : (x.b < 0.0 ? 0./0. : result.b);
   result.a = isNaN.a ? x.a : (x.a < 0.0 ? 0./0. : result.a);
   return result;
-`, sI = se({ opSnippet: eI, packedOpSnippet: tI, cpuKernelImpl: Zu }), nI = {
+`, sI = se({ opSnippet: eI, packedOpSnippet: tI, cpuKernelImpl: Ju }), nI = {
   kernelName: Rc,
   backendName: "webgl",
   kernelFunc: sI
@@ -8940,7 +8940,7 @@ const SI = hn + `
 `, NI = Ce({
   opSnippet: SI,
   packedOpSnippet: wI,
-  cpuKernelImpl: Ju
+  cpuKernelImpl: Zu
 }), RI = {
   kernelName: Tc,
   backendName: "webgl",
@@ -8950,7 +8950,7 @@ function TI(a) {
   const { inputs: t, backend: e, attrs: n } = a, { x: s } = t;
   At(s, "maxPool");
   const { filterSize: o, strides: r, pad: c, dimRoundingMode: i } = n, l = 1;
-  te(Ze(r, l), () => `Error in maxPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
+  te(Je(r, l), () => `Error in maxPool: Either strides or dilations must be 1. Got strides ${r} and dilations '${l}'`);
   const u = Le(s.shape, o, r, l, c, i);
   if (u.filterWidth === 1 && u.filterHeight === 1 && nt(u.inShape, u.outShape))
     return Fe({ inputs: { x: s }, backend: e });
@@ -8963,7 +8963,7 @@ const EI = {
   kernelFunc: TI
 };
 function DI(a) {
-  const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { filterSize: o, strides: r, pad: c, dataFormat: i, dimRoundingMode: l } = n, u = [1, 1, 1], p = Je(s.shape, o, r, u, c, l, i), d = new bn(p, "max", !1);
+  const { inputs: t, backend: e, attrs: n } = a, { x: s } = t, { filterSize: o, strides: r, pad: c, dataFormat: i, dimRoundingMode: l } = n, u = [1, 1, 1], p = Ze(s.shape, o, r, u, c, l, i), d = new bn(p, "max", !1);
   return e.runWebGLProgram(d, [s], s.dtype);
 }
 const FI = {
@@ -9096,7 +9096,7 @@ class OI {
   }
 }
 function AI(a) {
-  const { inputs: t, backend: e, attrs: n } = a, { dy: s, input: o } = t, r = o, { filterSize: c, strides: i, pad: l, dimRoundingMode: u } = n, p = [1, 1, 1], d = Je(r.shape, c, i, p, l, u), h = new bn(
+  const { inputs: t, backend: e, attrs: n } = a, { dy: s, input: o } = t, r = o, { filterSize: c, strides: i, pad: l, dimRoundingMode: u } = n, p = [1, 1, 1], d = Ze(r.shape, c, i, p, l, u), h = new bn(
     d,
     "max",
     !0
@@ -9134,7 +9134,7 @@ const MI = {
     const { x: n } = a, { filterSize: s, strides: o, pad: r, includeBatchInIndex: c } = t, i = e;
     te(n.shape.length === 4, () => `Error in maxPool: input must be rank 4 but got rank ${n.shape.length}.`);
     const l = [1, 1];
-    te(Ze(o, l), () => `Error in maxPool: Either strides or dilations must be 1. Got strides ${o} and dilations '${l}'`);
+    te(Je(o, l), () => `Error in maxPool: Either strides or dilations must be 1. Got strides ${o} and dilations '${l}'`);
     const u = Le(n.shape, s, o, l, r), [p, d] = WI(n, c, u, i);
     return [p, d];
   }
@@ -9332,19 +9332,19 @@ class YI {
 const QI = ({ inputs: a, backend: t, attrs: e }) => {
   const { x: n } = a, { paddings: s, mode: o } = e, r = _().getBool("WEBGL_PACK_ARRAY_OPERATIONS") ? new YI(n.shape, s, o) : new qI(n.shape, s, o);
   return t.runWebGLProgram(r, [n], n.dtype);
-}, ZI = {
+}, JI = {
   kernelName: fa,
   backendName: "webgl",
   kernelFunc: QI
 };
-const JI = `if (b == 0.0) return NAN;
+const ZI = `if (b == 0.0) return NAN;
   return mod(a, b);`, ey = `
   vec4 result = mod(a, b);
   bvec4 isNaN = equal(b, vec4(0.0));
   ` + it + `
   return result;
 `, ty = Ce({
-  opSnippet: JI,
+  opSnippet: ZI,
   packedOpSnippet: ey
 }), sy = {
   kernelName: js,
@@ -9767,7 +9767,7 @@ const Ky = We + `
 };
 const Qy = We + `
   return (x < 0.0) ? 0.0 : min(6.0, x);
-`, Zy = `
+`, Jy = `
   vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));
   bvec4 isNaN = isnan(x);
 
@@ -9777,10 +9777,10 @@ const Qy = We + `
   result.a = isNaN.a ? x.a : result.a;
 
   return result;
-`, Jy = se({ opSnippet: Qy, packedOpSnippet: Zy }), ek = {
+`, Zy = se({ opSnippet: Qy, packedOpSnippet: Jy }), ek = {
   kernelName: Rs,
   backendName: "webgl",
-  kernelFunc: Jy
+  kernelFunc: Zy
 };
 class tk {
   constructor(t, e, n, s, o) {
@@ -10576,7 +10576,7 @@ const _k = `
   float scale = ${Ar};
   return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);
 `, Lk = se({ opSnippet: _k }), Vk = {
-  kernelName: Zs,
+  kernelName: Js,
   backendName: "webgl",
   kernelFunc: Lk
 };
@@ -10605,7 +10605,7 @@ const zk = `
   if (isnan(x)) { return 0.0; }
   return sign(x);
 `, Uk = se({ opSnippet: zk }), Hk = {
-  kernelName: Js,
+  kernelName: Zs,
   backendName: "webgl",
   kernelFunc: Uk
 };
@@ -10624,12 +10624,12 @@ const Xk = It + `
 const Yk = `
   float e2x = exp(x);
   return (e2x - 1.0 / e2x) / 2.0;
-`, Qk = se({ opSnippet: Yk }), Zk = {
+`, Qk = se({ opSnippet: Yk }), Jk = {
   kernelName: tn,
   backendName: "webgl",
   kernelFunc: Qk
 };
-const Jk = `
+const Zk = `
   float epsilon = 1.1920928955078125e-7;
   float threshold = log(epsilon) + 2.0;
 
@@ -10649,7 +10649,7 @@ const Jk = `
     result = log(exp_x + 1.0);
   }
   return result;
-`, e$ = se({ opSnippet: Jk }), t$ = {
+`, e$ = se({ opSnippet: Zk }), t$ = {
   kernelName: sn,
   backendName: "webgl",
   kernelFunc: e$
@@ -10978,7 +10978,7 @@ function Mi(a) {
   return e.runWebGLProgram(r, [s], s.dtype);
 }
 const X$ = {
-  kernelName: Za,
+  kernelName: Ja,
   backendName: "webgl",
   kernelFunc: Mi
 };
@@ -11141,7 +11141,7 @@ function q$(a) {
   return $ = U({ inputs: { x: $ }, attrs: { shape: v }, backend: e }), tt(e, w), [$, I];
 }
 const Y$ = {
-  kernelName: Ja,
+  kernelName: Za,
   backendName: "webgl",
   kernelFunc: q$
 };
@@ -11282,7 +11282,7 @@ class Q$ {
         `;
   }
 }
-function Z$(a) {
+function J$(a) {
   const { inputs: t, backend: e, attrs: n } = a, { image: s, transforms: o } = t, { interpolation: r, fillMode: c, fillValue: i, outputShape: l } = n, [u, p, d, h] = s.shape, [f, m] = l ?? [p, d], x = [
     u,
     f,
@@ -11291,10 +11291,10 @@ function Z$(a) {
   ], g = new Q$(p, d, r, c, i, x);
   return e.runWebGLProgram(g, [s, o], "float32");
 }
-const J$ = {
+const Z$ = {
   kernelName: er,
   backendName: "webgl",
-  kernelFunc: Z$
+  kernelFunc: J$
 };
 function ev(a) {
   const { inputs: t, attrs: e, backend: n } = a, { axis: s } = e, { x: o } = t;
@@ -11460,7 +11460,7 @@ function av(a) {
   const d = Ql(p.shape, l, r), h = G([p.shape[l]]), f = U({ inputs: { x: p }, backend: e, attrs: { shape: [-1, h] } });
   i.push(f);
   const m = cr(s.dtype), x = (b, y, k, $, v) => {
-    const w = b.shape[0], S = b.shape[1], R = Zl(S, v), P = { windowSize: R, inSize: S, batchSize: w, numSegments: v }, D = new ov(P, y), F = e.compileAndRun(D, [b, k], $);
+    const w = b.shape[0], S = b.shape[1], R = Jl(S, v), P = { windowSize: R, inSize: S, batchSize: w, numSegments: v }, D = new ov(P, y), F = e.compileAndRun(D, [b, k], $);
     if (i.push(F), F.shape[1] === v)
       return F;
     const T = Wi({
@@ -11543,7 +11543,7 @@ const iv = [
   X1,
   q1,
   Y1,
-  Z1,
+  J1,
   e0,
   s0,
   o0,
@@ -11565,7 +11565,7 @@ const iv = [
   Sd,
   K0,
   Q0,
-  J0,
+  Z0,
   nI,
   rI,
   uI,
@@ -11583,7 +11583,7 @@ const iv = [
   GI,
   UI,
   jI,
-  ZI,
+  JI,
   sy,
   iy,
   Xx,
@@ -11624,7 +11624,7 @@ const iv = [
   Gk,
   Hk,
   qk,
-  Zk,
+  Jk,
   tb,
   ay,
   t$,
@@ -11651,7 +11651,7 @@ const iv = [
   z$,
   X$,
   Y$,
-  J$,
+  Z$,
   Kx,
   tv,
   nv,
@@ -11695,6 +11695,8 @@ class AS extends Ui {
   tokens = [];
   lastLoss = null;
   lastMultinomialRand = null;
+  jobQueue = [];
+  processingJob = !1;
   async tokenisePrompt(t, e, n) {
     if (e) {
       const s = e.length > 0 && e[e.length - 1].role === "assistant";
@@ -11856,7 +11858,7 @@ class AS extends Ui {
       this.cache = s, this.lastToken = -1;
     }
     const n = this.tokeniser.trained ? this.tokeniser : new Ad(uv(lv, this.tokeniser.vocabSize));
-    this.actualTokeniser = n;
+    this.actualTokeniser = n, e?.loraName ? this.model.attachLoRA(e.loraName) : this.model.hasLoRA() && this.model.detachLoRA();
   }
   async step(t, e) {
     const n = { ...e, maxLength: 1 };
@@ -11864,9 +11866,32 @@ class AS extends Ui {
   }
   async generate(t, e) {
     let n;
-    Array.isArray(t) ? n = t : typeof t == "object" && (e = t), this.initialise(n, e), this.active = !0, e?.maxLength !== 1 && this.emit("start");
-    const o = await this._generate(e);
-    return this.active = !1, this.emit("stop"), o;
+    if (Array.isArray(t) ? n = t : typeof t == "object" && (e = t), this.processingJob) {
+      if (this.jobQueue.length > 10)
+        throw new Error("Job queue is too long, rejecting new job");
+      return new Promise((s, o) => {
+        this.jobQueue.push({ prompt: n, options: e, resolve: s, reject: o });
+      });
+    }
+    this.processingJob = !0;
+    try {
+      const s = await this.startJob(n, e);
+      if (this.processingJob = !1, this.jobQueue.length > 0) {
+        const o = this.jobQueue.shift();
+        this.generate(o.prompt || [], o.options).then(o.resolve).catch(o.reject);
+      }
+      return s;
+    } catch (s) {
+      throw this.processingJob = !1, s;
+    }
+  }
+  async startJob(t, e) {
+    this.initialise(t, e), this.active = !0, e?.maxLength !== 1 && this.emit("start");
+    const s = await this._generate(e);
+    return this.active = !1, this.emit("stop"), s;
+  }
+  getQueueLength() {
+    return this.jobQueue.length;
   }
   stop() {
     this.active = !1;