@genai-fi/nanogpt 0.8.4 → 0.9.0

package/dist/ops/cpu/adamMoments.js

@@ -1,5 +1,5 @@
-import { f as p } from "../../index-DdmHGZjq.js";
-import { s as f } from "../../stack-BNwLzE43.js";
+import { f as p } from "../../index-BzFyqcy-.js";
+import { s as f } from "../../stack-DFatutCx.js";
 function b(t) {
   const { moments: n, gradient: c } = t.inputs, { beta1: o, beta2: m } = t.attrs, e = n.shape.length, a = new Array(e).fill(0), s = n.shape.slice();
   s[e - 1] = 1;

package/dist/ops/cpu/appendCache.js

@@ -1,5 +1,5 @@
-import { f as d } from "../../index-DdmHGZjq.js";
-import { c as h } from "../../concat-pHiVqR3L.js";
+import { f as d } from "../../index-BzFyqcy-.js";
+import { c as h } from "../../concat-B912vBbo.js";
 function u(p) {
   const { cache: n, item: s } = p.inputs, { maxSize: i, pastLen: c } = p.attrs, t = n.shape[0], o = n.shape[1], a = n.shape[3], e = s.shape[2];
   if (c + e <= i) {

package/dist/ops/cpu/attentionMask.js

@@ -1,8 +1,8 @@
-import { f as a, h as p, b as u } from "../../index-DdmHGZjq.js";
-import { l as N, w as b } from "../../ops-542ai2vG.js";
-import { o as A } from "../../ones-BAqVh-eA.js";
-import { z as I } from "../../zeros-Dnwix0p4.js";
-import { m as g } from "../../mat_mul-Dpy2mMRu.js";
+import { f as a, h as p, b as u } from "../../index-BzFyqcy-.js";
+import { l as N, w as b } from "../../ops-LuCMAnmM.js";
+import { o as A } from "../../ones-tIJeHlq-.js";
+import { z as I } from "../../zeros-Bj5rMYA7.js";
+import { m as g } from "../../mat_mul-DzjTFx-u.js";
 function o(n) {
   const { q: s, k: e } = n.inputs, { divisor: r } = n.attrs, c = s.shape[2], t = e.shape[2], m = N.bandPart(A([t, t]), -1, 0).cast("bool"), l = I([t, t]), i = p([t, t], Number.NEGATIVE_INFINITY), f = b(m, l, i), k = g(s, e, !1, !0).mul(u(r)), d = f.slice([0, 0], [c, t]).expandDims(0).expandDims(0);
   return k.add(d);

package/dist/ops/cpu/fusedSoftmax.js

@@ -1,5 +1,5 @@
-import { f as e } from "../../index-DdmHGZjq.js";
-import { s as f } from "../../softmax-BfsyI4As.js";
+import { f as e } from "../../index-BzFyqcy-.js";
+import { s as f } from "../../softmax-D7Jj3p_P.js";
 function n(t) {
   const { inputs: s, attrs: a } = t, { logits: o } = s, { dim: i, dropoutRate: r } = a;
   if (!o)

package/dist/ops/cpu/gatherSub.js

@@ -1,6 +1,6 @@
-import { C as u, D as c, E as g, a8 as h, f as m, c as p } from "../../index-DdmHGZjq.js";
-import { r as f } from "../../range-BcUvLuf5.js";
-import { s as l } from "../../stack-BNwLzE43.js";
+import { E as u, F as c, G as g, a8 as h, f as m, c as p } from "../../index-BzFyqcy-.js";
+import { r as f } from "../../range-CWcz7xFA.js";
+import { s as l } from "../../stack-DFatutCx.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -26,9 +26,9 @@ function d(e) {
   const { values: t, labels: n, logits: r } = e.inputs, s = n.shape[0], a = f(0, s, 1, "int32"), i = l([a, n], 1), o = b(r, i);
   return p(t, o);
 }
-const D = {
+const k = {
   kernelName: "EfficientGatherSub",
   backendName: "cpu",
   kernelFunc: d
 };
-m(D);
+m(k);

package/dist/ops/cpu/gelu.js

@@ -1,4 +1,4 @@
-import { f as t, t as d } from "../../index-DdmHGZjq.js";
+import { f as t, t as d } from "../../index-BzFyqcy-.js";
 const o = 0.7978845608028654, c = 0.044715;
 function m(r) {
   const { inputs: u } = r, { x: n } = u, e = n;

package/dist/ops/cpu/matMulGelu.js

@@ -1,5 +1,5 @@
-import { f as e, t as m } from "../../index-DdmHGZjq.js";
-import { g as d, d as M } from "../../gelu-BkcmEEyD.js";
+import { f as e, t as m } from "../../index-BzFyqcy-.js";
+import { g as d, d as M } from "../../gelu-Bp_-935b.js";
 function c(t) {
   const { inputs: u } = t, { x: n, kernel: l } = u, a = n, r = l;
   return m(() => {

package/dist/ops/cpu/matMulMul.js

@@ -1,4 +1,4 @@
-import { f as e, t as i } from "../../index-DdmHGZjq.js";
+import { f as e, t as i } from "../../index-BzFyqcy-.js";
 function n(t) {
   const { inputs: r, attrs: l } = t, { transposeA: o, transposeB: s } = l, { x: u, kernel: a, y: c } = r, m = u, k = a, M = c;
   return i(() => m.matMul(k, o, s).mul(M));

package/dist/ops/cpu/mulDropout.js

@@ -1,4 +1,4 @@
-import { f as e, a as t } from "../../index-DdmHGZjq.js";
+import { f as e, a as t } from "../../index-BzFyqcy-.js";
 function n(o) {
   const { inputs: r } = o, { a: l, b: u } = r;
   return console.warn("Using fallback mulDrop implementation without dropout."), t(l, u);

package/dist/ops/cpu/normRMS.js

@@ -1,4 +1,4 @@
-import { f as o, t as d } from "../../index-DdmHGZjq.js";
+import { f as o, t as d } from "../../index-BzFyqcy-.js";
 function i(t) {
   const { inputs: e } = t, { x: n, gamma: s } = e, r = n, a = s;
   return d(() => {

package/dist/ops/cpu/qkv.js

@@ -1,6 +1,6 @@
-import { f as q } from "../../index-DdmHGZjq.js";
-import { r as o } from "../../reshape-WeJkT3ja.js";
-import { s as x } from "../../split-BPxr8_8m.js";
+import { f as q } from "../../index-BzFyqcy-.js";
+import { r as o } from "../../reshape-CnIwVG1c.js";
+import { s as x } from "../../split-DK2k5eHf.js";
 function v(p) {
   const { x: c, kernel: K } = p.inputs, { heads: n } = p.attrs, [s, e, t] = c.shape, a = o(c, [s * e, t]), i = a.dot(K);
   a.dispose();

package/dist/ops/cpu/rope.js

@@ -1,8 +1,8 @@
-import { f as S } from "../../index-DdmHGZjq.js";
-import { r as F } from "../../range-BcUvLuf5.js";
-import { g as I } from "../../gather-CPg6ZlQA.js";
-import { s as E } from "../../stack-BNwLzE43.js";
-import { c as T } from "../../concat-pHiVqR3L.js";
+import { f as S } from "../../index-BzFyqcy-.js";
+import { r as F } from "../../range-CWcz7xFA.js";
+import { g as I } from "../../gather-Dnpgw-YQ.js";
+import { s as E } from "../../stack-DFatutCx.js";
+import { c as T } from "../../concat-B912vBbo.js";
 function U(t, c, p, o, r) {
   const n = o.shape[3], s = p;
   if (s > n) return o;

package/dist/ops/cpu/scatterSub.js

@@ -1,8 +1,8 @@
-import { C as f, M as g, D as r, E as l, a7 as N, f as b, c as S, a as D } from "../../index-DdmHGZjq.js";
-import { v as h } from "../../scatter_nd_util-B7yDhiQr.js";
-import { r as k } from "../../range-BcUvLuf5.js";
-import { s as v } from "../../stack-BNwLzE43.js";
-import { o as E } from "../../ones-BAqVh-eA.js";
+import { E as f, M as g, F as r, G as l, a7 as N, f as b, c as S, a as h } from "../../index-BzFyqcy-.js";
+import { v as D } from "../../scatter_nd_util-BQdz--Gn.js";
+import { r as k } from "../../range-CWcz7xFA.js";
+import { s as v } from "../../stack-DFatutCx.js";
+import { o as E } from "../../ones-tIJeHlq-.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -22,18 +22,18 @@ import { o as E } from "../../ones-BAqVh-eA.js";
 function I(a, e, s) {
   g(s);
   const n = r(a, "indices", "scatterND", "int32"), t = r(e, "updates", "scatterND");
-  h(t, n, s);
+  D(t, n, s);
   const c = { indices: n, updates: t }, o = { shape: s };
   return l.runKernel(N, c, o);
 }
-const C = /* @__PURE__ */ f({ scatterND_: I });
-function K(a) {
-  const { logits: e, labels: s, dy: n } = a.inputs, t = s.shape[0], c = e.shape[1], o = k(0, t, 1, "int32"), i = v([o, s], 1), d = E([t]), u = C(i, d, [t, c]), p = S(e, u), m = n.reshape([t, 1]);
-  return D(p, m);
+const K = /* @__PURE__ */ f({ scatterND_: I });
+function L(a) {
+  const { logits: e, labels: s, dy: n } = a.inputs, t = s.shape[0], c = e.shape[1], o = k(0, t, 1, "int32"), i = v([o, s], 1), d = E([t]), u = K(i, d, [t, c]), p = S(e, u), m = n.reshape([t, 1]);
+  return h(p, m);
 }
-const L = {
+const T = {
   kernelName: "EfficientScatterSub",
   backendName: "cpu",
-  kernelFunc: K
+  kernelFunc: L
 };
-b(L);
+b(T);

package/dist/ops/fusedSoftmax.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-DdmHGZjq.js";
+import { e as t } from "../index-BzFyqcy-.js";
 import "./cpu/fusedSoftmax.js";
 import "./webgl/fusedSoftmax.js";
 import "./grads/fusedSoftmax.js";

package/dist/ops/gatherSub.js

@@ -1,4 +1,4 @@
-import { e as n } from "../index-DdmHGZjq.js";
+import { e as n } from "../index-BzFyqcy-.js";
 import "./cpu/gatherSub.js";
 import "./webgl/gatherSub.js";
 function f(r, e, t) {

package/dist/ops/gelu.js

@@ -1,7 +1,7 @@
-import "../index-DdmHGZjq.js";
+import "../index-BzFyqcy-.js";
 import "./cpu/gelu.js";
 import "./webgl/gelu.js";
-import { d as e, g as i } from "../gelu-BkcmEEyD.js";
+import { d as e, g as i } from "../gelu-Bp_-935b.js";
 export {
   e as dGelu,
   i as gelu

package/dist/ops/grads/attentionMask.js

@@ -1,4 +1,4 @@
-import { i as m, b as i } from "../../index-DdmHGZjq.js";
+import { i as m, b as i } from "../../index-BzFyqcy-.js";
 import { matMulMul as u } from "../matMulMul.js";
 const p = {
   kernelName: "AttentionMask",

package/dist/ops/grads/fusedSoftmax.js

@@ -1,6 +1,6 @@
-import { i as f, a as i, c as l } from "../../index-DdmHGZjq.js";
+import { i as f, a as i, c as l } from "../../index-BzFyqcy-.js";
 import { mulDrop as g } from "../mulDrop.js";
-import { s as T } from "../../sum-ByFINZgi.js";
+import { s as T } from "../../sum-CJ0ULhmt.js";
 const Y = {
   kernelName: "FusedSoftmax",
   outputsToSave: [!0],

package/dist/ops/grads/gelu.js

@@ -1,5 +1,5 @@
-import "../../index-DdmHGZjq.js";
-import { a as e } from "../../gelu-BkcmEEyD.js";
+import "../../index-BzFyqcy-.js";
+import { a as e } from "../../gelu-Bp_-935b.js";
 export {
   e as geluGradConfig
 };

package/dist/ops/grads/matMulGelu.js

@@ -1,4 +1,4 @@
-import { i as a, e as o } from "../../index-DdmHGZjq.js";
+import { i as a, e as o } from "../../index-BzFyqcy-.js";
 function i(e, n, r) {
   return o().runKernel("MatMulGeluGrad", { dy: e, x: n, kernel: r });
 }

package/dist/ops/grads/normRMS.js

@@ -1,4 +1,4 @@
-import { i as t, e as g } from "../../index-DdmHGZjq.js";
+import { i as t, e as g } from "../../index-BzFyqcy-.js";
 function i(r, a, n) {
   return g().runKernel("RMSNormGrad", { dy: r, x: a, gamma: n });
 }

package/dist/ops/grads/qkv.js

@@ -1,4 +1,4 @@
-import { i as Q } from "../../index-DdmHGZjq.js";
+import { i as Q } from "../../index-BzFyqcy-.js";
 const V = {
   kernelName: "QKV",
   inputsToSave: ["x", "kernel"],

package/dist/ops/grads/rope.js

@@ -1,4 +1,4 @@
-import { i, e as a } from "../../index-DdmHGZjq.js";
+import { i, e as a } from "../../index-BzFyqcy-.js";
 function p(n, e, s, o) {
   return a().runKernel("Rope", { x: n, sin: e, cos: s }, { pastLen: o });
 }

package/dist/ops/matMulGelu.js

@@ -1,4 +1,4 @@
-import { e as u } from "../index-DdmHGZjq.js";
+import { e as u } from "../index-BzFyqcy-.js";
 import "./cpu/matMulGelu.js";
 import "./webgl/matMulGelu.js";
 import "./grads/matMulGelu.js";

package/dist/ops/matMulMul.js

@@ -1,4 +1,4 @@
-import { e as u } from "../index-DdmHGZjq.js";
+import { e as u } from "../index-BzFyqcy-.js";
 import "./cpu/matMulMul.js";
 import "./webgl/matMulMul.js";
 function m(e, r, t, l = !1, n = !1) {

package/dist/ops/mulDrop.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-DdmHGZjq.js";
+import { e as t } from "../index-BzFyqcy-.js";
 import "./cpu/mulDropout.js";
 import "./webgl/mulDropout.js";
 function m(r, o, e, n) {

package/dist/ops/normRMS.js

@@ -1,4 +1,4 @@
-import { e as n } from "../index-DdmHGZjq.js";
+import { e as n } from "../index-BzFyqcy-.js";
 import "./cpu/normRMS.js";
 import "./webgl/normRMS.js";
 import "./grads/normRMS.js";

package/dist/ops/qkv.js

@@ -1,4 +1,4 @@
-import { e as o } from "../index-DdmHGZjq.js";
+import { e as o } from "../index-BzFyqcy-.js";
 import "./cpu/qkv.js";
 import "./webgl/qkv.js";
 import "./grads/qkv.js";

package/dist/ops/rope.js

@@ -1,8 +1,8 @@
-import { e as p } from "../index-DdmHGZjq.js";
-import "../random_width-DKGeiFuR.js";
-import "../register_all_kernels-Do9VvZmo.js";
+import { e as p } from "../index-BzFyqcy-.js";
+import "../random_width-CXVRloNK.js";
+import "../register_all_kernels-DIGpEwcf.js";
 import "../index-Tf7vU29b.js";
-import "../dataset-DPPl-iLT.js";
+import "../dataset-DlZtKmBq.js";
 import "./cpu/rope.js";
 import "./webgl/rope.js";
 import "./grads/rope.js";

package/dist/ops/scatterSub.js

@@ -1,4 +1,4 @@
-import { e as i } from "../index-DdmHGZjq.js";
+import { e as i } from "../index-BzFyqcy-.js";
 import "./cpu/scatterSub.js";
 import "./webgl/scatterSub.js";
 function c(t, r, e) {

package/dist/ops/webgl/adamAdjust.js

@@ -1,5 +1,5 @@
-import { r as n } from "../../Reshape-Bh_jzKzV.js";
-import { f } from "../../index-DdmHGZjq.js";
+import { r as n } from "../../Reshape-Bowtk9BP.js";
+import { f } from "../../index-BzFyqcy-.js";
 class v {
   variableNames = ["moments", "value"];
   outputShape;

package/dist/ops/webgl/adamMoments.js

@@ -1,4 +1,4 @@
-import { f as m } from "../../index-DdmHGZjq.js";
+import { f as m } from "../../index-BzFyqcy-.js";
 class i {
   variableNames = ["moments", "gradient"];
   outputShape;

package/dist/ops/webgl/appendCache.js

@@ -1,4 +1,4 @@
-import { f as p } from "../../index-DdmHGZjq.js";
+import { f as p } from "../../index-BzFyqcy-.js";
 class m {
   variableNames = ["cache", "item"];
   outputShape;

package/dist/ops/webgl/attentionMask.js

@@ -1,4 +1,4 @@
-import { f as m } from "../../index-DdmHGZjq.js";
+import { f as m } from "../../index-BzFyqcy-.js";
 class h {
   variableNames = ["q", "k"];
   outputShape;

package/dist/ops/webgl/fusedSoftmax.js

@@ -1,7 +1,7 @@
-import { m as b, s as I, r as k } from "../../RealDiv-D_q39E3A.js";
-import { r as v } from "../../Reshape-Bh_jzKzV.js";
-import { f as w, p as P } from "../../index-DdmHGZjq.js";
-import { e as S } from "../../axis_util-Did9235A.js";
+import { m as b, s as I, r as k } from "../../RealDiv-D4EzDsC0.js";
+import { r as v } from "../../Reshape-Bowtk9BP.js";
+import { f as w, p as P } from "../../index-BzFyqcy-.js";
+import { e as S } from "../../axis_util-TbGYJ208.js";
 class T {
   variableNames = ["logits", "maxLogits"];
   outputShape;

package/dist/ops/webgl/gatherSub.js

@@ -1,4 +1,4 @@
-import { f as l } from "../../index-DdmHGZjq.js";
+import { f as l } from "../../index-BzFyqcy-.js";
 class u {
   variableNames = ["labels", "logits", "values"];
   outputShape;

package/dist/ops/webgl/gelu.js

@@ -1,17 +1,17 @@
-import { f as a } from "../../index-DdmHGZjq.js";
-import { u as s, C as c } from "../../kernel_funcs_utils-CDfFpUab.js";
-const t = 0.7978845608028654, r = 0.044715, d = c + `
+import { f as a } from "../../index-BzFyqcy-.js";
+import { u as s, C as i } from "../../kernel_funcs_utils-DKLK0Mg3.js";
+const t = 0.7978845608028654, r = 0.044715, c = i + `
     float x3 = x * x * x;
     float inner = x + ${r} * x3;
     inner = ${t} * inner;
-    inner = tanh(inner);
+    inner = abs(inner) > 15.0 ? sign(inner) : tanh(inner);
     inner = 0.5 * (1.0 + inner);
     inner = x * inner;
     return inner;
-`, i = s({ opSnippet: d }), x = {
+`, d = s({ opSnippet: c }), x = {
   kernelName: "Gelu",
   backendName: "webgl",
-  kernelFunc: i
+  kernelFunc: d
 };
 a(x);
 class f {
@@ -27,7 +27,7 @@ class f {
                 float x2 = x * x;
                 float x3 = x2 * x;
                 float u  = ${t} * (x + ${r} * x3);
-                float t  = tanh(u);
+                float t = abs(u) > 15.0 ? sign(u) : tanh(u);
                 float sech2 = 1.0 - t * t;
                 float du_dx = ${t} * (1.0 + 3.0 * ${r} * x2);
                 float dgelu = 0.5 * (1.0 + t) + 0.5 * x * sech2 * du_dx;
@@ -46,5 +46,5 @@ const p = {
 };
 a(p);
 export {
-  i as gelu
+  d as gelu
 };

package/dist/ops/webgl/log.js

@@ -1,6 +1,6 @@
-import { f as e, aj as r } from "../../index-DdmHGZjq.js";
-import { u as s, l as N } from "../../kernel_funcs_utils-CDfFpUab.js";
-import { y as l } from "../../shared-CZiWmQCI.js";
+import { f as e, aj as r } from "../../index-BzFyqcy-.js";
+import { u as s, l as N } from "../../kernel_funcs_utils-DKLK0Mg3.js";
+import { y as l } from "../../shared-DuP7ue-R.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.

package/dist/ops/webgl/matMulGelu.js

@@ -1,63 +1,73 @@
-import { f as _, t as R, e as C, j as A, l as N, n as H, u as O } from "../../index-DdmHGZjq.js";
-import { r as f } from "../../Reshape-Bh_jzKzV.js";
-import { M as U } from "../../mulmat_packed_gpu-q_Gmwyld.js";
-import { m as E } from "../../mat_mul-Dpy2mMRu.js";
-const M = 0.7978845608028654, x = 0.044715, j = `
+import { f as E, t as R, e as C, j as $, l as N, n as H, u as O } from "../../index-BzFyqcy-.js";
+import { r as f } from "../../Reshape-Bowtk9BP.js";
+import { M as U } from "../../mulmat_packed_gpu-BT60jmzP.js";
+import { m as A } from "../../mat_mul-DzjTFx-u.js";
+const M = 0.7978845608028654, g = 0.044715, j = `
     vec4 x3 = x * x * x;
-    vec4 inner = x + ${x} * x3;
+    vec4 inner = x + ${g} * x3;
     inner = ${M} * inner;
-    inner = tanh(inner);
+    inner = vec4(
+        abs(inner[0]) > 15.0 ? sign(inner[0]) : tanh(inner[0]),
+        abs(inner[1]) > 15.0 ? sign(inner[1]) : tanh(inner[1]),
+        abs(inner[2]) > 15.0 ? sign(inner[2]) : tanh(inner[2]),
+        abs(inner[3]) > 15.0 ? sign(inner[3]) : tanh(inner[3])
+    );
     inner = 0.5 * (1.0 + inner);
     vec4 result = x * inner;
     return result;
 `, q = `
     vec4 a2 = a * a;
     vec4 a3 = a2 * a;
-    vec4 u  = ${M} * (a + ${x} * a3);
-    vec4 t  = tanh(u);
+    vec4 u  = ${M} * (a + ${g} * a3);
+    vec4 t = vec4(
+        abs(u[0]) > 15.0 ? sign(u[0]) : tanh(u[0]),
+        abs(u[1]) > 15.0 ? sign(u[1]) : tanh(u[1]),
+        abs(u[2]) > 15.0 ? sign(u[2]) : tanh(u[2]),
+        abs(u[3]) > 15.0 ? sign(u[3]) : tanh(u[3])
+    );
     vec4 sech2 = 1.0 - t * t;
-    vec4 du_dx = ${M} * (1.0 + 3.0 * ${x} * a2);
+    vec4 du_dx = ${M} * (1.0 + 3.0 * ${g} * a2);
     vec4 dgelu = 0.5 * (1.0 + t) + 0.5 * a * sech2 * du_dx;
     return dgelu * b;
-`, te = 1e3;
-function w({
+`, ne = 1e3;
+function _({
   a: e,
-  b: t,
+  b: n,
   transposeA: s,
-  transposeB: n,
+  transposeB: t,
   backend: a,
   activationSnippet: c,
   multiplier: o
 }) {
-  const r = e.shape.length, u = t.shape.length, i = s ? e.shape[r - 2] : e.shape[r - 1], p = n ? t.shape[u - 1] : t.shape[u - 2], h = s ? e.shape[r - 1] : e.shape[r - 2], l = n ? t.shape[u - 2] : t.shape[u - 1], K = e.shape.slice(0, -2), T = t.shape.slice(0, -2), d = A(K), m = A(T), b = N(e.shape.slice(0, -2), t.shape.slice(0, -2)).concat([h, l]);
+  const r = e.shape.length, i = n.shape.length, u = s ? e.shape[r - 2] : e.shape[r - 1], h = t ? n.shape[i - 1] : n.shape[i - 2], p = s ? e.shape[r - 1] : e.shape[r - 2], l = t ? n.shape[i - 2] : n.shape[i - 1], w = e.shape.slice(0, -2), K = n.shape.slice(0, -2), d = $(w), m = $(K), T = N(e.shape.slice(0, -2), n.shape.slice(0, -2)).concat([p, l]);
   H(
-    i === p,
-    () => `Error in matMul: inner shapes (${i}) and (${p}) of Tensors with shapes ${e.shape} and ${t.shape} and transposeA=${s} and transposeB=${n} must match.`
+    u === h,
+    () => `Error in matMul: inner shapes (${u}) and (${h}) of Tensors with shapes ${e.shape} and ${n.shape} and transposeA=${s} and transposeB=${t} must match.`
   );
-  const v = s ? [d, i, h] : [d, h, i], S = n ? [m, l, p] : [m, p, l], D = f({ inputs: { x: e }, backend: a, attrs: { shape: v } }), G = f({ inputs: { x: t }, backend: a, attrs: { shape: S } }), g = [D, G], y = Math.max(d, m), L = c, B = O(e.dtype, t.dtype), F = new U(
+  const v = s ? [d, u, p] : [d, p, u], x = t ? [m, l, h] : [m, h, l], S = f({ inputs: { x: e }, backend: a, attrs: { shape: v } }), b = f({ inputs: { x: n }, backend: a, attrs: { shape: x } }), D = [S, b], y = Math.max(d, m), L = c, B = O(e.dtype, n.dtype), F = new U(
     v,
-    S,
-    [y, h, l],
+    x,
+    [y, p, l],
     s,
-    n,
+    t,
     !1,
     L,
     !!o,
     !1
-  ), k = [D, G];
-  o && k.push(o);
-  const $ = a.runWebGLProgram(F, k, B), I = f({ inputs: { x: $ }, backend: a, attrs: { shape: b } });
-  g.push($);
-  for (const P of g)
+  ), G = [S, b];
+  o && G.push(o);
+  const k = a.runWebGLProgram(F, G, B), I = f({ inputs: { x: k }, backend: a, attrs: { shape: T } });
+  D.push(k);
+  for (const P of D)
     a.disposeIntermediateTensorInfo(P);
   return I;
 }
 function z(e) {
-  const { inputs: t, backend: s } = e, { x: n, kernel: a } = t;
-  if (n === void 0 || a === void 0)
+  const { inputs: n, backend: s } = e, { x: t, kernel: a } = n;
+  if (t === void 0 || a === void 0)
     throw new Error("BatchMatMul requires two input tensors.");
-  return w({
-    a: n,
+  return _({
+    a: t,
     b: a,
     transposeA: !1,
     transposeB: !1,
@@ -70,21 +80,21 @@ const W = {
   backendName: "webgl",
   kernelFunc: z
 };
-_(W);
+E(W);
 function J(e) {
-  const { dy: t, x: s, kernel: n } = e.inputs, a = e.backend;
+  const { dy: n, x: s, kernel: t } = e.inputs, a = e.backend;
   return R(() => {
     const c = C().makeTensorFromTensorInfo(
-      w({
+      _({
         a: s,
-        b: n,
+        b: t,
         transposeA: !1,
         transposeB: !1,
         backend: a,
         activationSnippet: q,
-        multiplier: t
+        multiplier: n
       })
-    ), o = E(c, n, !1, !0), r = E(s, c, !0, !1);
+    ), o = A(c, t, !1, !0), r = A(s, c, !0, !1);
     return [o, r];
   });
 }
@@ -93,9 +103,9 @@ const Q = {
   backendName: "webgl",
   kernelFunc: J
 };
-_(Q);
+E(Q);
 export {
-  te as MATMUL_SHARED_DIM_THRESHOLD,
-  w as batchMatMulGeluImpl,
+  ne as MATMUL_SHARED_DIM_THRESHOLD,
+  _ as batchMatMulGeluImpl,
   z as batchMatMulKernel
 };

package/dist/ops/webgl/matMulMul.js

@@ -1,4 +1,4 @@
-import { f as u } from "../../index-DdmHGZjq.js";
+import { f as u } from "../../index-BzFyqcy-.js";
 import { batchMatMulGeluImpl as c } from "./matMulGelu.js";
 const M = `
     return a * b;

package/dist/ops/webgl/mulDropout.js

@@ -1,4 +1,4 @@
-import { f as m } from "../../index-DdmHGZjq.js";
+import { f as m } from "../../index-BzFyqcy-.js";
 class f {
   variableNames = ["a", "b"];
   outputShape;

package/dist/ops/webgl/normRMS.js

@@ -1,5 +1,5 @@
-import { f as p, e as G } from "../../index-DdmHGZjq.js";
-import { s as x } from "../../sum-ByFINZgi.js";
+import { f as p, e as G } from "../../index-BzFyqcy-.js";
+import { s as x } from "../../sum-CJ0ULhmt.js";
 class y {
   variableNames = ["x", "meanSquare", "gamma"];
   outputShape;

package/dist/ops/webgl/qkv.js

@@ -1,4 +1,4 @@
-import { f as i } from "../../index-DdmHGZjq.js";
+import { f as i } from "../../index-BzFyqcy-.js";
 class l {
   variableNames = ["x", "kernel"];
   outputShape;

package/dist/ops/webgl/rope.js

@@ -1,4 +1,4 @@
-import { f as u } from "../../index-DdmHGZjq.js";
+import { f as u } from "../../index-BzFyqcy-.js";
 class l {
   variableNames = ["x", "sin", "cos"];
   outputShape;

package/dist/ops/webgl/scatterSub.js

@@ -1,4 +1,4 @@
-import { f as i } from "../../index-DdmHGZjq.js";
+import { f as i } from "../../index-BzFyqcy-.js";
 class u {
   variableNames = ["labels", "softmaxProbs", "dy"];
   outputShape;

package/dist/ops/webgpu/adamAdjust.js

@@ -1,6 +1,6 @@
-import { g as p } from "../../webgpu_program-Dhk9R5aG.js";
-import { f as d, c as l } from "../../webgpu_util-BqGnZg8t.js";
-import { f, a4 as c } from "../../index-DdmHGZjq.js";
+import { g as p } from "../../webgpu_program-DkQJOJSd.js";
+import { f as d, c as l } from "../../webgpu_util-pLEV9tks.js";
+import { f, a4 as c } from "../../index-BzFyqcy-.js";
 class h {
   variableNames = ["moments", "value"];
   outputShape;

package/dist/ops/webgpu/adamMoments.js

@@ -1,6 +1,6 @@
-import { g as m } from "../../webgpu_program-Dhk9R5aG.js";
-import { f as u, c as p } from "../../webgpu_util-BqGnZg8t.js";
-import { f as d, a4 as c } from "../../index-DdmHGZjq.js";
+import { g as m } from "../../webgpu_program-DkQJOJSd.js";
+import { f as u, c as p } from "../../webgpu_util-pLEV9tks.js";
+import { f as d, a4 as c } from "../../index-BzFyqcy-.js";
 class h {
   variableNames = ["moments", "gradient"];
   outputShape;