@genai-fi/nanogpt 0.17.5 → 0.18.0

package/dist/ops/webgpu/transpose16.js

@@ -1,13 +1,13 @@
 import { isPackedTensor as d } from "../../utilities/packed.js";
-import { c as k, h as l } from "../../index-CUXkjxiT.js";
+import { a as k, e as l } from "../../index-DSGwv2Yx.js";
 import { reshape16 as u } from "../reshape16.js";
 import P from "./transpose16_shared_program.js";
 import b from "./transpose16_program.js";
-import { t as T } from "../../transpose-9kRxIXWR.js";
-function w(p) {
-  const { inputs: h, attrs: i } = p, { x: e } = h, { perm: r } = i, a = p.backend, c = d(e);
-  if (c && r[r.length - 1] !== e.shape.length - 1) {
-    const o = e.shape.length, m = o === 4 ? r.map((s) => s - 1).filter((s) => s >= 0) : r, t = o === 4 ? u(e, [e.shape[0] * e.shape[1], e.shape[2], e.shape[3]]) : e, f = new P(t.shape, m), n = a.runWebGPUProgram(f, [t], "packedF16");
+import { t as T } from "../../transpose-Qxz-4os3.js";
+function w(a) {
+  const { inputs: h, attrs: i } = a, { x: e } = h, { perm: r } = i, p = a.backend, m = d(e);
+  if (m && r[r.length - 1] !== e.shape.length - 1) {
+    const o = e.shape.length, c = o === 4 ? r.map((s) => s - 1).filter((s) => s >= 0) : r, t = o === 4 ? u(e, [e.shape[0] * e.shape[1], e.shape[2], e.shape[3]]) : e, f = new P(t.shape, c), n = p.runWebGPUProgram(f, [t], "packedF16");
     if (o === 4) {
       t.dispose();
       const s = l().makeTensorFromTensorInfo(n), g = u(s, [
@@ -20,9 +20,9 @@ function w(p) {
     }
     return n;
   }
-  if (c) {
+  if (m) {
     const o = new b(e.shape, r);
-    return a.runWebGPUProgram(o, [e], "packedF16");
+    return p.runWebGPUProgram(o, [e], "packedF16");
   } else
     return T(e, r);
 }

package/dist/ops/webgpu/transpose16_program.js

@@ -1,5 +1,5 @@
-import { f as a, c as i } from "../../webgpu_util-DYlGSwOJ.js";
-import { b as h, e as d, c as n } from "../../webgpu_program-B4HmApL1.js";
+import { f as a, c as i } from "../../webgpu_util-DuofJBMo.js";
+import { b as h, e as d, c as n } from "../../webgpu_program-CbjdYLYk.js";
 function p(r) {
   const e = r.length;
   if (e > 6)

package/dist/ops/webgpu/transpose16_shared_program.js

@@ -1,6 +1,6 @@
-import { c as a } from "../../webgpu_util-DYlGSwOJ.js";
-import { e as l } from "../../webgpu_program-B4HmApL1.js";
-import { x as p } from "../../index-CUXkjxiT.js";
+import { c as a } from "../../webgpu_util-DuofJBMo.js";
+import { e as l } from "../../webgpu_program-CbjdYLYk.js";
+import { v as p } from "../../index-DSGwv2Yx.js";
 class n {
   variableNames = ["A"];
   outputShape;

package/dist/ops/webgpu/unpack16.js

@@ -1,6 +1,6 @@
-import { f as c, c as r } from "../../webgpu_util-DYlGSwOJ.js";
-import { e as u } from "../../webgpu_program-B4HmApL1.js";
-import { c as p } from "../../index-CUXkjxiT.js";
+import { f as c, c as r } from "../../webgpu_util-DuofJBMo.js";
+import { e as u } from "../../webgpu_program-CbjdYLYk.js";
+import { a as p } from "../../index-DSGwv2Yx.js";
 class l {
   outputShape;
   shaderKey = "Unpack16";

package/dist/ops/webgpu/utils/binary_op.js

@@ -1,8 +1,8 @@
-import { f as r, c as n } from "../../../webgpu_util-DYlGSwOJ.js";
+import { f as r, c as n } from "../../../webgpu_util-DuofJBMo.js";
 import { g as p } from "../../../binary_op_util-pKXltfxI.js";
 import { B as k } from "../../../binary_op_util-pKXltfxI.js";
-import { _ as c, ai as s } from "../../../index-CUXkjxiT.js";
-import { e as u } from "../../../webgpu_program-B4HmApL1.js";
+import { _ as c, ai as s } from "../../../index-DSGwv2Yx.js";
+import { e as u } from "../../../webgpu_program-CbjdYLYk.js";
 class x {
   dispatch;
   dispatchLayout;

package/dist/ops/webgpu/utils/reductions.js

@@ -1,9 +1,9 @@
-import { ah as h, U as S, h as f } from "../../../index-CUXkjxiT.js";
-import { e as d } from "../../../webgpu_program-B4HmApL1.js";
+import { ah as h, N as S, e as f } from "../../../index-DSGwv2Yx.js";
+import { e as d } from "../../../webgpu_program-CbjdYLYk.js";
 import { reshape16 as g } from "../../reshape16.js";
-import { f as z } from "../../../webgpu_util-DYlGSwOJ.js";
-import { c as k } from "../../../axis_util-GTVlo58H.js";
-import { z as x } from "../../../zeros-DvZpK8s6.js";
+import { f as z } from "../../../webgpu_util-DuofJBMo.js";
+import { c as k } from "../../../axis_util-QWWgLjut.js";
+import { z as x } from "../../../zeros-Bw0puq_w.js";
 function c(e, u, t, i) {
   return e && !u ? `
             bestValue = subgroupAdd(bestValue);

package/dist/{ops-CsXeTq1P.js → ops-CURIZSVt.js}

@@ -1,38 +1,38 @@
-import { o as h, q as u, x as p, E as k, bd as ds, be as hs, a5 as bs, bf as gs, bg as Es, L as W, bh as $s, bi as ys, bj as _s, bk as Ns, a as $, m as d, j as G, i as S, f as Ss, a1 as b, bl as ks, w as K, d as Is, a6 as L, a_ as ns, n as Ts, a2 as Ms } from "./index-CUXkjxiT.js";
-import { t as v } from "./tile-CFy-xTO6.js";
-import { e as R } from "./expand_dims-Bkd1YD5x.js";
-import { t as x } from "./tensor1d-LMGMIUlr.js";
+import { o as d, n as u, v as p, E as k, bd as hs, be as ds, a5 as bs, bf as gs, bg as Es, J as W, bh as $s, bi as ys, bj as _s, bk as Ns, b as $, m as h, h as G, d as S, f as Ss, a1 as b, bl as ks, u as K, l as Is, a6 as L, a_ as ns, j as Ts, a2 as Ms } from "./index-DSGwv2Yx.js";
+import { t as j } from "./tile-CsWlVKKz.js";
+import { e as R } from "./expand_dims-D0rBtgT1.js";
+import { t as x } from "./tensor1d-BMl0eZYV.js";
 import { n as xs, a as As, b as ws } from "./non_max_suppression_impl-B2W7YjZB.js";
-import { r as Os, l as ls, g as es, a as rs, m as Q, b as Gs, c as Ls, u as us, n as ps, s as Bs, d as Ds, e as Ws, f as Rs } from "./resize_nearest_neighbor-BojqlfRe.js";
-import { s as ms } from "./split-C2Sj255c.js";
-import { s as _ } from "./sum-BpiwSWvg.js";
-import { w as q, a as qs, n as zs, b as as, e as ss, l as Vs } from "./not_equal-hurPF26l.js";
-import { s as T } from "./slice-BzS11Qh0.js";
-import { r as F } from "./range-C-CjF-LI.js";
-import { t as Cs } from "./tensor-BWFldCso.js";
-import { n as C, t as is } from "./transpose-9kRxIXWR.js";
-import { r as O } from "./reshape-BDOuCSNW.js";
-import { s as U } from "./stack-DudVrtmG.js";
-import { z as Ps } from "./zeros-DvZpK8s6.js";
-import { s as Ys } from "./squeeze-ho4wLUek.js";
-import { c as X } from "./concat-CAQpCret.js";
-import { m as w } from "./mat_mul-DhG0Newp.js";
-import { t as J } from "./tensor2d-BnXMKScO.js";
-import { o as js } from "./ones-BytntneX.js";
-import { r as fs } from "./relu-J_X6MUzx.js";
-import { e as vs } from "./axis_util-GTVlo58H.js";
+import { r as Os, l as ls, g as es, a as rs, m as Q, b as Gs, c as Ls, u as us, n as ps, s as Bs, d as Ds, e as Ws, f as Rs } from "./resize_nearest_neighbor-C6_0dAnK.js";
+import { s as ms } from "./split-BN9LkEgS.js";
+import { s as _ } from "./sum-D39FeU5h.js";
+import { w as q, a as qs, n as zs, b as as, e as ss, l as Vs } from "./not_equal-CkQKkKZy.js";
+import { s as T } from "./slice-BvItlgXu.js";
+import { r as F } from "./range-BDxO73mk.js";
+import { t as Cs } from "./tensor-D8e0Gd7c.js";
+import { n as C, t as is } from "./transpose-Qxz-4os3.js";
+import { r as O } from "./reshape-BIN71H3p.js";
+import { s as U } from "./stack-z6QE7kmP.js";
+import { z as vs } from "./zeros-Bw0puq_w.js";
+import { s as Ps } from "./squeeze-O_YWJpw_.js";
+import { c as J } from "./concat-Dqk7Xk7h.js";
+import { m as w } from "./mat_mul-DP86qZtZ.js";
+import { t as X } from "./tensor2d-DTtQ1QcT.js";
+import { o as Ys } from "./ones-DbVB5N58.js";
+import { r as fs } from "./relu-DTvZKBsZ.js";
+import { e as js } from "./axis_util-QWWgLjut.js";
 function Fs(t, a, n) {
   const r = u(t, "x", "bincount"), e = u(a, "weights", "bincount");
   p(r.dtype === "int32", () => `Error in bincount: input dtype must be int32, but got ${r.dtype}`), p(n >= 0, () => `size must be non-negative, but got ${n}.`), p(e.size === r.size || e.size === 0, () => `Error in bincount: weights must have the same size as input or0-length, but got input shape: ${r.shape}, weights shape: ${e.shape}.`);
   const s = { x: r, weights: e }, o = { size: n };
-  return k.runKernel(ds, s, o);
+  return k.runKernel(hs, s, o);
 }
-const Us = /* @__PURE__ */ h({ bincount_: Fs });
+const Us = /* @__PURE__ */ d({ bincount_: Fs });
 function Zs(t, ...a) {
   const n = a.map((e, s) => u(e, `tensors${s}`, "einsum")), r = { equation: t };
-  return k.runKernel(hs, n, r);
+  return k.runKernel(ds, n, r);
 }
-const V = /* @__PURE__ */ h({ einsum_: Zs });
+const V = /* @__PURE__ */ d({ einsum_: Zs });
 function Hs(t, a, n, r = "float32") {
   a == null && (a = t);
   const e = bs([t, a], r), s = t <= a ? t : a;
@@ -42,11 +42,11 @@ function Hs(t, a, n, r = "float32") {
   if (n == null)
     return o;
   if (n.length === 1)
-    return v(R(o, 0), [n[0], 1, 1]);
+    return j(R(o, 0), [n[0], 1, 1]);
   if (n.length === 2)
-    return v(R(R(o, 0), 0), [n[0], n[1], 1, 1]);
+    return j(R(R(o, 0), 0), [n[0], n[1], 1, 1]);
   if (n.length === 3)
-    return v(R(R(R(o, 0), 0), 0), [
+    return j(R(R(R(o, 0), 0), 0), [
       n[0],
       n[1],
       n[2],
@@ -55,28 +55,28 @@ function Hs(t, a, n, r = "float32") {
     ]);
   throw new Error(`eye() currently supports only 1D and 2D batchShapes, but received ${n.length}D.`);
 }
-const Ks = /* @__PURE__ */ h({ eye_: Hs });
-function Xs(t, a, n, r, e = "bilinear", s = 0) {
+const Ks = /* @__PURE__ */ d({ eye_: Hs });
+function Js(t, a, n, r, e = "bilinear", s = 0) {
   const o = u(t, "image", "cropAndResize"), i = u(a, "boxes", "cropAndResize", "float32"), l = u(n, "boxInd", "cropAndResize", "int32"), c = i.shape[0];
   p(o.rank === 4, () => `Error in cropAndResize: image must be rank 4,but got rank ${o.rank}.`), p(i.rank === 2 && i.shape[1] === 4, () => `Error in cropAndResize: boxes must be have size [${c},4] but had shape ${i.shape}.`), p(l.rank === 1 && l.shape[0] === c, () => `Error in cropAndResize: boxInd must be have size [${c}] but had shape ${i.shape}.`), p(r.length === 2, () => `Error in cropAndResize: cropSize must be of length 2, but got length ${r.length}.`), p(r[0] >= 1 && r[1] >= 1, () => `cropSize must be atleast [1,1], but was ${r}`), p(e === "bilinear" || e === "nearest", () => `method must be bilinear or nearest, but was ${e}`);
   const f = { image: o, boxes: i, boxInd: l }, m = { method: e, extrapolationValue: s, cropSize: r };
   return k.runKernel(gs, f, m);
 }
-const Js = /* @__PURE__ */ h({ cropAndResize_: Xs });
+const Xs = /* @__PURE__ */ d({ cropAndResize_: Js });
 function Qs(t) {
   const a = u(t, "image", "flipLeftRight", "float32");
   p(a.rank === 4, () => `Error in flipLeftRight: image must be rank 4,but got rank ${a.rank}.`);
   const n = { image: a };
   return k.runKernel(Es, n, {});
 }
-const sn = /* @__PURE__ */ h({ flipLeftRight_: Qs });
+const sn = /* @__PURE__ */ d({ flipLeftRight_: Qs });
 function nn(t) {
   const a = u(t, "image", "grayscaleToRGB"), n = a.rank - 1, r = a.shape[n];
   p(a.rank >= 2, () => `Error in grayscaleToRGB: images must be at least rank 2, but got rank ${a.rank}.`), p(r === 1, () => `Error in grayscaleToRGB: last dimension of a grayscale image should be size 1, but got size ${r}.`);
   const e = new Array(a.rank);
-  return e.fill(1, 0, n), e[n] = 3, v(a, e);
+  return e.fill(1, 0, n), e[n] = 3, j(a, e);
 }
-const en = /* @__PURE__ */ h({ grayscaleToRGB_: nn });
+const en = /* @__PURE__ */ d({ grayscaleToRGB_: nn });
 function tn(t) {
   const a = u(t, "image", "RGBToGrayscale"), n = a.rank - 1, r = a.shape[n];
   p(a.rank >= 2, () => `Error in RGBToGrayscale: images must be at least rank 2, but got rank ${a.rank}.`), p(r === 3, () => `Error in RGBToGrayscale: last dimension of an RGB image should be size 3, but got size ${r}.`);
@@ -103,14 +103,14 @@ function tn(t) {
   }
   return i = R(i, -1), W(i, e);
 }
-const on = /* @__PURE__ */ h({ rgbToGrayscale_: tn });
+const on = /* @__PURE__ */ d({ rgbToGrayscale_: tn });
 function rn(t, a, n = 0, r = 0.5) {
   const e = u(t, "image", "rotateWithOffset", "float32");
   p(e.rank === 4, () => `Error in rotateWithOffset: image must be rank 4,but got rank ${e.rank}.`);
   const s = { image: e }, o = { radians: a, fillValue: n, center: r };
   return k.runKernel($s, s, o);
 }
-const an = /* @__PURE__ */ h({ rotateWithOffset_: rn });
+const an = /* @__PURE__ */ d({ rotateWithOffset_: rn });
 function z(t, a, n, r, e, s) {
   r == null && (r = 0.5), e == null && (e = Number.NEGATIVE_INFINITY), s == null && (s = 0);
   const o = t.shape[0];
@@ -122,7 +122,7 @@ function cn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY) {
   const l = { maxOutputSize: n, iouThreshold: r, scoreThreshold: e };
   return k.runKernel(ys, { boxes: s, scores: o }, l);
 }
-const ln = /* @__PURE__ */ h({ nonMaxSuppression_: cn });
+const ln = /* @__PURE__ */ d({ nonMaxSuppression_: cn });
 async function un(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY) {
   const s = u(t, "boxes", "nonMaxSuppressionAsync"), o = u(a, "scores", "nonMaxSuppressionAsync"), i = z(s, o, n, r, e);
   n = i.maxOutputSize, r = i.iouThreshold, e = i.scoreThreshold;
@@ -136,8 +136,8 @@ function mn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = 0) {
   const c = { boxes: o, scores: i }, f = { maxOutputSize: n, iouThreshold: r, scoreThreshold: e, softNmsSigma: s }, m = k.runKernel(_s, c, f);
   return { selectedIndices: m[0], selectedScores: m[1] };
 }
-const fn = /* @__PURE__ */ h({ nonMaxSuppressionWithScore_: mn });
-async function dn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = 0) {
+const fn = /* @__PURE__ */ d({ nonMaxSuppressionWithScore_: mn });
+async function hn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = 0) {
   const o = u(t, "boxes", "nonMaxSuppressionAsync"), i = u(a, "scores", "nonMaxSuppressionAsync"), l = z(o, i, n, r, e, s);
   n = l.maxOutputSize, r = l.iouThreshold, e = l.scoreThreshold, s = l.softNmsSigma;
   const c = await Promise.all([o.data(), i.data()]), f = c[0], m = c[1], { selectedIndices: E, selectedScores: g } = As(f, m, n, r, e, s);
@@ -146,7 +146,7 @@ async function dn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = 0) {
     selectedScores: x(g)
   };
 }
-const hn = dn;
+const dn = hn;
 function bn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = !1) {
   const o = u(t, "boxes", "nonMaxSuppression"), i = u(a, "scores", "nonMaxSuppression"), l = z(
     o,
@@ -164,7 +164,7 @@ function bn(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = !1) {
   }, y = k.runKernel(Ns, E, g);
   return { selectedIndices: y[0], validOutputs: y[1] };
 }
-const gn = /* @__PURE__ */ h({ nonMaxSuppressionPadded_: bn });
+const gn = /* @__PURE__ */ d({ nonMaxSuppressionPadded_: bn });
 async function En(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = !1) {
   const o = u(t, "boxes", "nonMaxSuppressionAsync"), i = u(a, "scores", "nonMaxSuppressionAsync"), l = z(
     o,
@@ -183,10 +183,10 @@ async function En(t, a, n, r = 0.5, e = Number.NEGATIVE_INFINITY, s = !1) {
 const $n = En;
 function yn(t, a = "binary", n = !1, r = 0.5) {
   const e = u(t, "image", "threshold"), s = 0.2989, o = 0.587, i = 0.114, l = e.shape[0] * e.shape[1];
-  let c = d(x([r]), 255), f, m, E, g;
+  let c = h(x([r]), 255), f, m, E, g;
   if (p(e.rank === 3, () => `Error in threshold: image must be rank 3,but got rank ${e.rank}.`), p(e.shape[2] === 3 || e.shape[2] === 1, () => `Error in threshold: image color channel must be equal to 3 or 1but got ${e.shape[2]}.`), p(e.dtype === "int32" || e.dtype === "float32", () => `Error in dtype: image dtype must be int32 or float32,but got dtype ${e.dtype}.`), p(a === "otsu" || a === "binary", () => `Method must be binary or otsu, but was ${a}`), e.shape[2] === 3) {
     [f, m, E] = ms(e, [1, 1, 1], -1);
-    const I = d(f, s), A = d(m, o), D = d(E, i);
+    const I = h(f, s), A = h(m, o), D = h(E, i);
     g = G(G(I, A), D);
   } else
     g = t;
@@ -195,41 +195,41 @@ function yn(t, a = "binary", n = !1, r = 0.5) {
     c = _n(I, l);
   }
   const y = n ? ls(g, c) : es(g, c);
-  return W(d(y, 255), "int32");
+  return W(h(y, 255), "int32");
 }
 function _n(t, a) {
   let n = x([-1]), r = x([0]), e = x([0]), s, o, i, l, c, f;
   for (let m = 0; m < t.size - 1; m++) {
     s = T(t, 0, m + 1), o = T(t, m + 1), c = S(_(s), a), f = S(_(o), a);
-    const E = _(d(s, F(0, s.size)));
+    const E = _(h(s, F(0, s.size)));
     i = S(E, _(s));
-    const g = Ss(o.shape, s.size), y = G(F(0, o.size), g), M = d(o, y);
+    const g = Ss(o.shape, s.size), y = G(F(0, o.size), g), M = h(o, y);
     l = S(_(M), _(o));
-    const I = b(i, l), A = b(i, l), D = d(c, f);
-    e = d(d(D, I), A);
-    const P = es(e, r);
-    r = q(P, e, r), n = q(P, x([m]), n);
+    const I = b(i, l), A = b(i, l), D = h(c, f);
+    e = h(h(D, I), A);
+    const v = es(e, r);
+    r = q(v, e, r), n = q(v, x([m]), n);
   }
   return n;
 }
-const Nn = /* @__PURE__ */ h({ threshold_: yn });
+const Nn = /* @__PURE__ */ d({ threshold_: yn });
 function Sn(t, a, n = "nearest", r = "constant", e = 0, s) {
   const o = u(t, "image", "transform", "float32"), i = u(a, "transforms", "transform", "float32");
   p(o.rank === 4, () => `Error in transform: image must be rank 4,but got rank ${o.rank}.`), p(i.rank === 2 && (i.shape[0] === o.shape[0] || i.shape[0] === 1) && i.shape[1] === 8, () => "Error in transform: Input transform should be batch x 8 or 1 x 8"), p(s == null || s.length === 2, () => `Error in transform: outputShape must be [height, width] or null, but got ${s}.`);
   const l = { image: o, transforms: i }, c = { interpolation: n, fillMode: r, fillValue: e, outputShape: s };
   return k.runKernel(ks, l, c);
 }
-const kn = /* @__PURE__ */ h({ transform_: Sn });
+const kn = /* @__PURE__ */ d({ transform_: Sn });
 function In(t, a, n) {
   const r = u(t, "a", "bandPart");
   p(r.rank >= 2, () => `bandPart(): Rank must be at least 2, got ${r.rank}.`);
   const e = r.shape, [s, o] = r.shape.slice(-2);
   let i, l;
   typeof a == "number" ? (p(a % 1 === 0, () => `bandPart(): numLower must be an integer, got ${a}.`), p(a <= s, () => `bandPart(): numLower (${a}) must not be greater than the number of rows (${s}).`), i = u(a < 0 ? s : a, "numLower", "bandPart")) : (p(a.dtype === "int32", () => "bandPart(): numLower's dtype must be an int32."), i = q(rs(a, 0), s, Q(a, s))), typeof n == "number" ? (p(n % 1 === 0, () => `bandPart(): numUpper must be an integer, got ${n}.`), p(n <= o, () => `bandPart(): numUpper (${n}) must not be greater than the number of columns (${o}).`), l = u(n < 0 ? o : n, "numUpper", "bandPart")) : (p(n.dtype === "int32", () => "bandPart(): numUpper's dtype must be an int32."), l = q(rs(n, 0), o, Q(n, o)));
-  const c = O(F(0, s, 1, "int32"), [-1, 1]), f = F(0, o, 1, "int32"), m = b(c, f), E = Gs(ls(m, i), Ls(m, C(l))), g = Ps([s, o], r.dtype);
+  const c = O(F(0, s, 1, "int32"), [-1, 1]), f = F(0, o, 1, "int32"), m = b(c, f), E = Gs(ls(m, i), Ls(m, C(l))), g = vs([s, o], r.dtype);
   return O(U(us(O(r, [-1, s, o])).map((y) => q(E, y, g))), e);
 }
-const Tn = /* @__PURE__ */ h({ bandPart_: In });
+const Tn = /* @__PURE__ */ d({ bandPart_: In });
 function Mn(t) {
   let a;
   if (Array.isArray(t)) {
@@ -238,7 +238,7 @@ function Mn(t) {
     for (let s = 1; s < t.length; ++s)
       p(t[s].shape[0] === e, () => `Gram-Schmidt: Non-unique lengths found in the input vectors: (${t[s].shape[0]} vs. ${e})`);
   } else
-    a = !0, t = ms(t, t.shape[0], 0).map((e) => Ys(e, [0]));
+    a = !0, t = ms(t, t.shape[0], 0).map((e) => Ps(e, [0]));
   p(t.length <= t[0].shape[0], () => `Gram-Schmidt: Number of vectors (${t.length}) exceeds number of dimensions (${t[0].shape[0]}).`);
   const n = [], r = t;
   for (let e = 0; e < t.length; ++e)
@@ -246,14 +246,14 @@ function Mn(t) {
       let s = r[e];
       if (e > 0)
         for (let o = 0; o < e; ++o) {
-          const i = d(_(d(n[o], s)), n[o]);
+          const i = h(_(h(n[o], s)), n[o]);
           s = b(s, i);
         }
       return S(s, ps(s, "euclidean"));
     }));
   return a ? U(n, 0) : n;
 }
-const xn = /* @__PURE__ */ h({ gramSchmidt_: Mn });
+const xn = /* @__PURE__ */ d({ gramSchmidt_: Mn });
 function An(t, a = !1) {
   if (p(t.rank >= 2, () => `qr() requires input tensor to have a rank >= 2, but got rank ${t.rank}`), t.rank === 2)
     return cs(t, a);
@@ -276,30 +276,30 @@ function cs(t, a = !1) {
     p(t.shape.length === 2, () => `qr2d() requires a 2D Tensor, but got a ${t.shape.length}D Tensor.`);
     const n = t.shape[0], r = t.shape[1];
     let e = Ks(n), s = K(t);
-    const o = J([[1]], [1, 1]);
+    const o = X([[1]], [1, 1]);
     let i = K(o);
     const l = n >= r ? r : n;
     for (let c = 0; c < l; ++c) {
       const f = s, m = i, E = e;
       [i, s, e] = k.tidy(() => {
-        const g = T(s, [c, c], [n - c, 1]), y = ps(g), M = T(s, [c, c], [1, 1]), I = q(es(M, 0), J([[-1]]), J([[1]])), A = b(M, d(I, y)), D = S(g, A);
-        D.shape[0] === 1 ? i = K(o) : i = X([
+        const g = T(s, [c, c], [n - c, 1]), y = ps(g), M = T(s, [c, c], [1, 1]), I = q(es(M, 0), X([[-1]]), X([[1]])), A = b(M, h(I, y)), D = S(g, A);
+        D.shape[0] === 1 ? i = K(o) : i = J([
           o,
           T(D, [1, 0], [D.shape[0] - 1, D.shape[1]])
         ], 0);
-        const P = C(S(w(I, A), y)), Y = T(s, [c, 0], [n - c, r]), Z = d(P, i), ts = is(i);
+        const v = C(S(w(I, A), y)), P = T(s, [c, 0], [n - c, r]), Z = h(v, i), ts = is(i);
         if (c === 0)
-          s = b(Y, w(Z, w(ts, Y)));
+          s = b(P, w(Z, w(ts, P)));
         else {
-          const H = b(Y, w(Z, w(ts, Y)));
-          s = X([T(s, [0, 0], [c, r]), H], 0);
+          const H = b(P, w(Z, w(ts, P)));
+          s = J([T(s, [0, 0], [c, r]), H], 0);
         }
-        const os = is(Z), j = T(e, [0, c], [n, e.shape[1] - c]);
+        const os = is(Z), Y = T(e, [0, c], [n, e.shape[1] - c]);
         if (c === 0)
-          e = b(j, w(w(j, i), os));
+          e = b(Y, w(w(Y, i), os));
         else {
-          const H = b(j, w(w(j, i), os));
-          e = X([T(e, [0, 0], [n, c]), H], 1);
+          const H = b(Y, w(w(Y, i), os));
+          e = J([T(e, [0, 0], [n, c]), H], 1);
         }
         return [i, s, e];
       }), Is([f, m, E]);
@@ -307,7 +307,7 @@ function cs(t, a = !1) {
     return !a && n > r && (e = T(e, [0, 0], [n, r]), s = T(s, [0, 0], [r, r])), [e, s];
   });
 }
-const wn = /* @__PURE__ */ h({ qr_: An });
+const wn = /* @__PURE__ */ d({ qr_: An });
 var N;
 (function(t) {
   t[t.NONE = 0] = "NONE", t[t.MEAN = 1] = "MEAN", t[t.SUM = 2] = "SUM", t[t.SUM_BY_NONZERO_WEIGHTS = 3] = "SUM_BY_NONZERO_WEIGHTS";
@@ -316,7 +316,7 @@ function On(t, a, n = N.SUM_BY_NONZERO_WEIGHTS) {
   const r = u(t, "losses", "computeWeightedLoss");
   let e = null;
   a != null && (e = u(a, "weights", "computeWeightedLoss"));
-  const s = e == null ? r : d(r, e);
+  const s = e == null ? r : h(r, e);
   if (n === N.NONE)
     return s;
   if (n === N.SUM)
@@ -333,13 +333,13 @@ function On(t, a, n = N.SUM_BY_NONZERO_WEIGHTS) {
     if (e == null)
       return S(_(s), $(r.size));
     {
-      const o = d(e, js(r.shape)), i = W(_(zs(o, $(0))), "float32");
+      const o = h(e, Ys(r.shape)), i = W(_(zs(o, $(0))), "float32");
       return S(_(s), i);
     }
   }
   throw Error(`Unknown reduction: ${n}`);
 }
-const B = /* @__PURE__ */ h({ computeWeightedLoss_: On });
+const B = /* @__PURE__ */ d({ computeWeightedLoss_: On });
 function Gn(t, a, n, r = N.SUM_BY_NONZERO_WEIGHTS) {
   const e = u(t, "labels", "absoluteDifference"), s = u(a, "predictions", "absoluteDifference");
   let o = null;
@@ -347,80 +347,80 @@ function Gn(t, a, n, r = N.SUM_BY_NONZERO_WEIGHTS) {
   const i = ns(b(e, s));
   return B(i, o, r);
 }
-const Ln = /* @__PURE__ */ h({ absoluteDifference_: Gn });
+const Ln = /* @__PURE__ */ d({ absoluteDifference_: Gn });
 function Bn(t, a, n, r, e = N.SUM_BY_NONZERO_WEIGHTS) {
   const s = u(t, "labels", "cosineDistance"), o = u(a, "predictions", "cosineDistance");
   let i = null;
   r != null && (i = u(r, "weights", "cosineDistance")), L(s.shape, o.shape, "Error in cosineDistance: ");
-  const l = $(1), c = b(l, _(d(s, o), n, !0));
+  const l = $(1), c = b(l, _(h(s, o), n, !0));
   return B(c, i, e);
 }
-const Dn = /* @__PURE__ */ h({ cosineDistance_: Bn });
+const Dn = /* @__PURE__ */ d({ cosineDistance_: Bn });
 function Wn(t, a, n, r = N.SUM_BY_NONZERO_WEIGHTS) {
   let e = u(t, "labels", "hingeLoss");
   const s = u(a, "predictions", "hingeLoss");
   let o = null;
   n != null && (o = u(n, "weights", "hingeLoss")), L(e.shape, s.shape, "Error in hingeLoss: ");
   const i = $(1);
-  e = b(d($(2), e), i);
-  const l = fs(b(i, d(e, s)));
+  e = b(h($(2), e), i);
+  const l = fs(b(i, h(e, s)));
   return B(l, o, r);
 }
-const Rn = /* @__PURE__ */ h({ hingeLoss_: Wn });
+const Rn = /* @__PURE__ */ d({ hingeLoss_: Wn });
 function qn(t, a, n, r = 1, e = N.SUM_BY_NONZERO_WEIGHTS) {
   const s = u(t, "labels", "huberLoss"), o = u(a, "predictions", "huberLoss");
   let i = null;
   n != null && (i = u(n, "weights", "huberLoss")), L(s.shape, o.shape, "Error in huberLoss: ");
-  const l = $(r), c = ns(b(o, s)), f = Q(c, l), m = b(c, f), E = G(d($(0.5), Ts(f)), d(l, m));
+  const l = $(r), c = ns(b(o, s)), f = Q(c, l), m = b(c, f), E = G(h($(0.5), Ts(f)), h(l, m));
   return B(E, i, e);
 }
-const zn = /* @__PURE__ */ h({ huberLoss_: qn });
+const zn = /* @__PURE__ */ d({ huberLoss_: qn });
 function Vn(t, a, n, r = 1e-7, e = N.SUM_BY_NONZERO_WEIGHTS) {
   const s = u(t, "labels", "logLoss"), o = u(a, "predictions", "logLoss");
   let i = null;
   n != null && (i = u(n, "weights", "logLoss")), L(s.shape, o.shape, "Error in logLoss: ");
-  const l = $(1), c = $(r), f = C(d(s, as(G(o, c)))), m = d(b(l, s), as(G(b(l, o), c))), E = b(f, m);
+  const l = $(1), c = $(r), f = C(h(s, as(G(o, c)))), m = h(b(l, s), as(G(b(l, o), c))), E = b(f, m);
   return B(E, i, e);
 }
-const Cn = /* @__PURE__ */ h({ logLoss_: Vn });
-function Pn(t, a, n, r = N.SUM_BY_NONZERO_WEIGHTS) {
+const Cn = /* @__PURE__ */ d({ logLoss_: Vn });
+function vn(t, a, n, r = N.SUM_BY_NONZERO_WEIGHTS) {
   const e = u(t, "labels", "meanSquaredError"), s = u(a, "predictions", "meanSquaredError");
   let o = null;
   n != null && (o = u(n, "weights", "meanSquaredError")), L(e.shape, s.shape, "Error in meanSquaredError: ");
   const i = Bs(e, s);
   return B(i, o, r);
 }
-const Yn = /* @__PURE__ */ h({ meanSquaredError_: Pn });
-function jn(t, a) {
+const Pn = /* @__PURE__ */ d({ meanSquaredError_: vn });
+function Yn(t, a) {
   const n = u(t, "labels", "sigmoidCrossEntropyWithLogits"), r = u(a, "logits", "sigmoidCrossEntropyWithLogits");
   L(n.shape, r.shape, "Error in sigmoidCrossEntropyWithLogits: ");
-  const e = fs(r), s = d(r, n), o = Ds(ss(C(ns(r))));
+  const e = fs(r), s = h(r, n), o = Ds(ss(C(ns(r))));
   return G(b(e, s), o);
 }
-function vn(t, a, n, r = 0, e = N.SUM_BY_NONZERO_WEIGHTS) {
+function jn(t, a, n, r = 0, e = N.SUM_BY_NONZERO_WEIGHTS) {
   let s = u(t, "multiClassLabels", "sigmoidCrossEntropy");
   const o = u(a, "logits", "sigmoidCrossEntropy");
   let i = null;
   if (n != null && (i = u(n, "weights", "sigmoidCrossEntropy")), L(s.shape, o.shape, "Error in sigmoidCrossEntropy: "), r > 0) {
     const c = $(r), f = $(1), m = $(0.5);
-    s = G(d(s, b(f, c)), d(m, c));
+    s = G(h(s, b(f, c)), h(m, c));
   }
-  const l = jn(s, o);
+  const l = Yn(s, o);
   return B(l, i, e);
 }
-const Fn = /* @__PURE__ */ h({ sigmoidCrossEntropy_: vn });
+const Fn = /* @__PURE__ */ d({ sigmoidCrossEntropy_: jn });
 function Un(t, a, n = -1) {
   if (n === -1 && (n = a.rank - 1), n !== a.rank - 1)
     throw Error(`Softmax cross entropy along a non-last dimension is not yet supported. Labels / logits was rank ${a.rank} and dim was ${n}`);
   return Ms((e, s, o) => {
     const l = Vs(s, [n], !0), c = b(W(s, "float32"), l);
     o([e, c]);
-    const f = C(d(c, e));
+    const f = C(h(c, e));
     return { value: _(f, [n]), gradFunc: (g, y) => {
-      const [M, I] = y, A = vs(g.shape, [n]);
+      const [M, I] = y, A = js(g.shape, [n]);
       return [
-        d(O(g, A), b(W(M, "float32"), ss(I))),
-        d(O(g, A), b(ss(I), W(M, "float32")))
+        h(O(g, A), b(W(M, "float32"), ss(I))),
+        h(O(g, A), b(ss(I), W(M, "float32")))
       ];
     } };
   })(t, a);
@@ -431,12 +431,12 @@ function Zn(t, a, n, r = 0, e = N.SUM_BY_NONZERO_WEIGHTS) {
   let i = null;
   if (n != null && (i = u(n, "weights", "softmaxCrossEntropy")), L(s.shape, o.shape, "Error in softmaxCrossEntropy: "), r > 0) {
     const c = $(r), f = $(1), m = $(s.shape[1]);
-    s = G(d(s, b(f, c)), S(c, m));
+    s = G(h(s, b(f, c)), S(c, m));
   }
   const l = Un(s, o);
   return B(l, i, e);
 }
-const Hn = /* @__PURE__ */ h({ softmaxCrossEntropy_: Zn });
+const Hn = /* @__PURE__ */ d({ softmaxCrossEntropy_: Zn });
 const $e = {
   flipLeftRight: sn,
   grayscaleToRGB: en,
@@ -444,11 +444,11 @@ const $e = {
   resizeBilinear: Rs,
   rgbToGrayscale: on,
   rotateWithOffset: an,
-  cropAndResize: Js,
+  cropAndResize: Xs,
   nonMaxSuppression: ln,
   nonMaxSuppressionAsync: pn,
   nonMaxSuppressionWithScore: fn,
-  nonMaxSuppressionWithScoreAsync: hn,
+  nonMaxSuppressionWithScoreAsync: dn,
   nonMaxSuppressionPadded: gn,
   nonMaxSuppressionPaddedAsync: $n,
   threshold: Nn,
@@ -464,7 +464,7 @@ const $e = {
   hingeLoss: Rn,
   huberLoss: zn,
   logLoss: Cn,
-  meanSquaredError: Yn,
+  meanSquaredError: Pn,
   sigmoidCrossEntropy: Fn,
   softmaxCrossEntropy: Hn
 };

package/dist/{pack16-bqltoUlR.js → pack16-WlOSOuZA.js}

@@ -1,6 +1,6 @@
-import { e as o, h as i } from "./index-CUXkjxiT.js";
+import { c as o, e as i } from "./index-DSGwv2Yx.js";
 import { isPackedTensor as t } from "./utilities/packed.js";
-import { s as u } from "./slice-BzS11Qh0.js";
+import { s as u } from "./slice-BvItlgXu.js";
 const s = {
   kernelName: "Unpack16",
   inputsToSave: [],

package/dist/patches/webgpu_backend.js

@@ -1,8 +1,8 @@
-import { U as c, af as l, ab as m, x as n } from "../index-CUXkjxiT.js";
-import "../webgpu_util-DYlGSwOJ.js";
-import { W as P } from "../backend_webgpu-BqASlsbV.js";
+import { N as c, af as l, ab as m, v as n } from "../index-DSGwv2Yx.js";
+import "../webgpu_util-DuofJBMo.js";
+import { W as P } from "../backend_webgpu-DI2wXEC2.js";
 import { compileProgram as y } from "./webgpu_program.js";
-import { m as M } from "../webgpu_program-B4HmApL1.js";
+import { m as M } from "../webgpu_program-CbjdYLYk.js";
 const b = (o, s) => {
   const i = o.limits.maxComputeWorkgroupsPerDimension, t = s.dispatchLayout, a = s.dispatch;
   if (a.every((r) => r <= i))
@@ -17,7 +17,7 @@ const b = (o, s) => {
     () => "Total dispatch size exceeds WebGPU maximum."
   ), [e, e, e]) : [e, e, 1];
 };
-class U extends P {
+class v extends P {
   subgroupMaxSize;
   subgroupMinSize;
   constructor(s, i) {
@@ -52,5 +52,5 @@ class U extends P {
   }
 }
 export {
-  U as default
+  v as default
 };

package/dist/patches/webgpu_base.js

@@ -1,4 +1,4 @@
-import { ae as n } from "../index-CUXkjxiT.js";
+import { ae as n } from "../index-DSGwv2Yx.js";
 import p from "./webgpu_backend.js";
 function c(a) {
   n(

package/dist/patches/webgpu_program.js

@@ -1,5 +1,5 @@
-import { ab as z, x as F, ac as O, ad as _ } from "../index-CUXkjxiT.js";
-import { g as A, d as b, a as x, b as g, s as L, t as l, c as y } from "../webgpu_program-B4HmApL1.js";
+import { ab as z, v as F, ac as O, ad as _ } from "../index-DSGwv2Yx.js";
+import { g as A, d as b, a as x, b as g, s as L, t as l, c as y } from "../webgpu_program-CbjdYLYk.js";
 var N = /* @__PURE__ */ ((s) => (s[s.FROM_PIXELS = 0] = "FROM_PIXELS", s[s.DRAW = 1] = "DRAW", s))(N || {});
 const H = (s, t, e, o, u) => {
   const a = { dtype: o.dtype, shape: o.shape }, n = D(e, a, t), r = s.createShaderModule({ code: n, label: t.constructor.name });

package/dist/{random_normal-IBRrha8a.js → random_normal-CIm8lk2-.js}

@@ -1,4 +1,4 @@
-import { o as u, u as m, a5 as f } from "./index-CUXkjxiT.js";
+import { o as u, q as m, a5 as f } from "./index-DSGwv2Yx.js";
 import { M as i } from "./rand_util-CZ7yLoUm.js";
 function c(n, s = 0, e = 1, o, t) {
   if (m(n), o != null && o === "bool")