@genai-fi/nanogpt 0.11.0 → 0.12.1

package/dist/ops/reshape16.js

@@ -1,5 +1,5 @@
-import { j as p, h as s, e as u } from "../index-D0RBWjq8.js";
-import { r as c } from "../reshape-CkjKPPqB.js";
+import { i as p, f as s, e as u } from "../index-Duu1Lvvv.js";
+import { r as c } from "../reshape-BI0yzp1T.js";
 const i = {
   kernelName: "Reshape16",
   inputsToSave: ["x"],
@@ -21,18 +21,18 @@ const l = {
   kernelFunc: a
 };
 s(l);
-const h = {
+const g = {
   kernelName: "Reshape16",
   backendName: "webgl",
   kernelFunc: a
 };
-s(h);
-const g = {
+s(g);
+const h = {
   kernelName: "Reshape16",
   backendName: "cpu",
   kernelFunc: a
 };
-s(g);
+s(h);
 function m(e, r) {
   return u().runKernel("Reshape16", { x: e }, { shape: r });
 }

package/dist/ops/rope.js

@@ -1,7 +1,7 @@
-import "../index-D0RBWjq8.js";
+import "../index-Duu1Lvvv.js";
 import "./cpu/rope.js";
 import "./webgl/rope.js";
-import { r as i } from "../rope-BmZmp9uP.js";
+import { r as i } from "../rope-DJ7Y7c-u.js";
 export {
   i as rope
 };

package/dist/ops/scatterSub.js

@@ -1,4 +1,4 @@
-import { e as i } from "../index-D0RBWjq8.js";
+import { e as i } from "../index-Duu1Lvvv.js";
 import "./cpu/scatterSub.js";
 import "./webgl/scatterSub.js";
 function c(t, r, e) {

package/dist/ops/slice16.js

@@ -1,6 +1,6 @@
 import { isPackedTensor as n } from "../utilities/packed.js";
-import { e as c } from "../index-D0RBWjq8.js";
-import { s as i } from "../slice-Aqy7KbJh.js";
+import { e as c } from "../index-Duu1Lvvv.js";
+import { s as i } from "../slice-wBNvzVyz.js";
 function a(r, e, o) {
   return n(r) ? c().runKernel("Slice16", { x: r }, { begin: e, size: o }) : i(r, e, o);
 }

package/dist/ops/softmax16.js

@@ -1,4 +1,4 @@
-import { e as n } from "../index-D0RBWjq8.js";
+import { e as n } from "../index-Duu1Lvvv.js";
 import "./grads/softmax16.js";
 import { isPackedTensor as e } from "../utilities/packed.js";
 function t(r) {

package/dist/ops/sub16.js

@@ -1,4 +1,4 @@
-import { c as s, e as t } from "../index-D0RBWjq8.js";
+import { c as s, e as t } from "../index-Duu1Lvvv.js";
 import { isPackedTensor as n } from "../utilities/packed.js";
 function c(r, e) {
   return !n(r) && !n(e) ? s(r, e) : t().runKernel("Sub16", { a: r, b: e });

package/dist/ops/sum16.js

@@ -1,6 +1,6 @@
-import { e as t } from "../index-D0RBWjq8.js";
+import { e as t } from "../index-Duu1Lvvv.js";
 import { isPackedTensor as s } from "../utilities/packed.js";
-import { s as n } from "../sum-BdplSvq_.js";
+import { s as n } from "../sum-BPUfDB2X.js";
 function p(r, o, e = !1) {
   if (!s(r))
     return n(r, o, e);

package/dist/ops/transpose16.js

@@ -1,20 +1,20 @@
-import { j as i, h as p, e as u } from "../index-D0RBWjq8.js";
+import { i, f as p, e as u } from "../index-Duu1Lvvv.js";
 import { forcePacked as l, forceFloat as m } from "./grads/utils.js";
-import { g } from "../axis_util-DofAuy0p.js";
-import { isPackedTensor as f } from "../utilities/packed.js";
-import { t as a } from "../transpose-JawVKyZy.js";
+import { g as f } from "../axis_util-DGqbT-FX.js";
+import { isPackedTensor as g } from "../utilities/packed.js";
+import { t as a } from "../transpose-BUkQCJp9.js";
 const d = {
   kernelName: "Transpose16",
   gradFunc: (r, s, t) => {
     if (Array.isArray(r))
       throw new Error("Transpose16 gradient does not support multiple outputs.");
-    const n = t, { perm: e } = n, o = g(e);
+    const n = t, { perm: e } = n, o = f(e);
     return { x: () => T(r, o) };
   }
 };
 i(d);
 function c(r) {
-  const { inputs: s, attrs: t } = r, { x: n } = s, { perm: e } = t, o = f(n);
+  const { inputs: s, attrs: t } = r, { x: n } = s, { perm: e } = t, o = g(n);
   if (o && e[e.length - 1] !== n.shape.length - 1)
     throw new Error("Transpose16 currently only supports the last axis being unchanged.");
   return o ? l(a(m(n), e)) : a(n, e);

package/dist/ops/unpack16.js

@@ -1,5 +1,5 @@
-import "../index-D0RBWjq8.js";
-import { u as t } from "../pack16-k4jq6aMX.js";
+import "../index-Duu1Lvvv.js";
+import { u as t } from "../pack16-atD0eYRm.js";
 import "../utilities/packed.js";
 export {
   t as unpack16

package/dist/ops/webgl/adamAdjust.js

@@ -1,5 +1,5 @@
-import { r as n } from "../../Reshape-Ct266DEk.js";
-import { h as f } from "../../index-D0RBWjq8.js";
+import { r as n } from "../../Reshape-Ck29jQSY.js";
+import { f } from "../../index-Duu1Lvvv.js";
 class v {
   variableNames = ["moments", "value"];
   outputShape;

package/dist/ops/webgl/adamMoments.js

@@ -1,4 +1,4 @@
-import { h as m } from "../../index-D0RBWjq8.js";
+import { f as m } from "../../index-Duu1Lvvv.js";
 class i {
   variableNames = ["moments", "gradient"];
   outputShape;

package/dist/ops/webgl/appendCache.js

@@ -1,4 +1,4 @@
-import { h as p } from "../../index-D0RBWjq8.js";
+import { f as p } from "../../index-Duu1Lvvv.js";
 class m {
   variableNames = ["cache", "item"];
   outputShape;

package/dist/ops/webgl/attentionMask.js

@@ -1,5 +1,5 @@
-import { h } from "../../index-D0RBWjq8.js";
-class m {
+import { f as m } from "../../index-Duu1Lvvv.js";
+class h {
   variableNames = ["q", "k"];
   outputShape;
   userCode;
@@ -34,7 +34,7 @@ class m {
   }
 }
 function l(o) {
-  const { q: t, k: e } = o.inputs, { divisor: s, pastLen: n } = o.attrs, a = o.backend, i = t.shape[0], r = t.shape[2], c = e.shape[2], u = t.shape[1], p = t.shape[3], d = new m(i, u, r, c, p);
+  const { q: t, k: e } = o.inputs, { divisor: s, pastLen: n } = o.attrs, a = o.backend, i = t.shape[0], r = t.shape[2], c = e.shape[2], u = t.shape[1], p = t.shape[3], d = new h(i, u, r, c, p);
   return a.runWebGLProgram(d, [t, e], "float32", [[s], [n], [Number.NEGATIVE_INFINITY]]);
 }
 const f = {
@@ -42,4 +42,4 @@ const f = {
   backendName: "webgl",
   kernelFunc: l
 };
-h(f);
+m(f);

package/dist/ops/webgl/fusedSoftmax.js

@@ -1,7 +1,7 @@
-import { m as b, s as I, r as k } from "../../RealDiv-Ds-jvL09.js";
-import { r as v } from "../../Reshape-Ct266DEk.js";
-import { h as w, af as P } from "../../index-D0RBWjq8.js";
-import { e as S } from "../../axis_util-DofAuy0p.js";
+import { m as b, s as I, r as k } from "../../RealDiv-C8neBwFi.js";
+import { r as v } from "../../Reshape-Ck29jQSY.js";
+import { f as w, af as P } from "../../index-Duu1Lvvv.js";
+import { e as S } from "../../axis_util-DGqbT-FX.js";
 class T {
   variableNames = ["logits", "maxLogits"];
   outputShape;
@@ -60,11 +60,11 @@ function L(r) {
   o.disposeIntermediateTensorInfo(d);
   const p = I({ inputs: { x: s }, backend: o, attrs: { axis: i, keepDims: !1 } }), a = v({ inputs: { x: p }, backend: o, attrs: { shape: f } });
   if (n !== void 0 && n > 0) {
-    const h = new C(e.shape), g = o.runWebGLProgram(h, [s, a], "float32", [
+    const g = new C(e.shape), h = o.runWebGLProgram(g, [s, a], "float32", [
       [n],
       [c ?? Math.random() * 1e4]
     ]);
-    return o.disposeIntermediateTensorInfo(s), o.disposeIntermediateTensorInfo(p), o.disposeIntermediateTensorInfo(a), g;
+    return o.disposeIntermediateTensorInfo(s), o.disposeIntermediateTensorInfo(p), o.disposeIntermediateTensorInfo(a), h;
   }
   const x = k({ inputs: { a: s, b: a }, backend: o });
   return o.disposeIntermediateTensorInfo(s), o.disposeIntermediateTensorInfo(p), o.disposeIntermediateTensorInfo(a), x;

package/dist/ops/webgl/gatherSub.js

@@ -1,4 +1,4 @@
-import { h as l } from "../../index-D0RBWjq8.js";
+import { f as l } from "../../index-Duu1Lvvv.js";
 class u {
   variableNames = ["labels", "logits", "values"];
   outputShape;

package/dist/ops/webgl/gelu.js

@@ -1,5 +1,5 @@
-import { h as a } from "../../index-D0RBWjq8.js";
-import { u as s, C as i } from "../../kernel_funcs_utils-CSaumNDs.js";
+import { f as a } from "../../index-Duu1Lvvv.js";
+import { u as s, C as i } from "../../kernel_funcs_utils-BtYrPoJu.js";
 const t = 0.7978845608028654, r = 0.044715, c = i + `
     float x3 = x * x * x;
     float inner = x + ${r} * x3;

package/dist/ops/webgl/log.js

@@ -1,6 +1,6 @@
-import { h as e, ao as r } from "../../index-D0RBWjq8.js";
-import { u as s, l as N } from "../../kernel_funcs_utils-CSaumNDs.js";
-import { y as l } from "../../shared-B7USJZgw.js";
+import { f as e, ao as r } from "../../index-Duu1Lvvv.js";
+import { u as s, l as N } from "../../kernel_funcs_utils-BtYrPoJu.js";
+import { y as l } from "../../shared-Q3BS6T03.js";
 const a = N + `
   return x < 0.0 ? NAN : log(x);
 `, t = `

package/dist/ops/webgl/matMul16.js

@@ -1,27 +1,27 @@
 import { isPackedTensor as k } from "../../utilities/packed.js";
-import { h as g, m as M, b as m } from "../../index-D0RBWjq8.js";
+import { f as g, m as M, b as m } from "../../index-Duu1Lvvv.js";
 import { matMulMul as N } from "../matMulMul.js";
 import { matMulGelu as U } from "../matMulGelu.js";
-import { m as G } from "../../mat_mul-DxpNTCRz.js";
-import { r as h } from "../../reshape-CkjKPPqB.js";
-import { t as w } from "../../transpose-JawVKyZy.js";
+import { m as G } from "../../mat_mul-Bn2BDpT4.js";
+import { r as w } from "../../reshape-BI0yzp1T.js";
+import { t as h } from "../../transpose-BUkQCJp9.js";
 function P(l) {
-  const { A: r, B: o } = l.inputs, { transposeA: p, transposeB: c, scale: u, activation: A, scaleA: d, scaleB: f, forceOutputShape: t, perm: n } = l.attrs, B = !k(r), v = !k(o);
+  const { A: r, B: o } = l.inputs, { transposeA: p, transposeB: c, scale: u, activation: A, scaleA: f, scaleB: d, forceOutputShape: t, perm: n } = l.attrs, B = !k(r), v = !k(o);
   if (B && v) {
-    const a = d !== void 0 ? M(r, m(d)) : r, i = f !== void 0 ? M(o, m(f)) : o;
+    const a = f !== void 0 ? M(r, m(f)) : r, i = d !== void 0 ? M(o, m(d)) : o;
     let e;
     if (u !== void 0 ? e = N(a, i, m(u), p, c) : A === "gelu" ? e = U(a, i) : e = G(a, i, p, c), n)
       if (t) {
-        const s = h(e, t);
+        const s = w(e, t);
         e.dispose();
-        const b = w(s, n);
+        const b = h(s, n);
         return s.dispose(), b;
       } else {
-        const s = w(e, n);
+        const s = h(e, n);
         return e.dispose(), s;
       }
     else if (t) {
-      const s = h(e, t);
+      const s = w(e, t);
       return e.dispose(), s;
     } else
       return e;

package/dist/ops/webgl/matMulGelu.js

@@ -1,7 +1,7 @@
-import "../../index-D0RBWjq8.js";
-import "../../Reshape-Ct266DEk.js";
-import { a as m, b as o, c as p } from "../../matMulGelu-B2s_80-H.js";
-import "../../mat_mul-DxpNTCRz.js";
+import "../../index-Duu1Lvvv.js";
+import "../../Reshape-Ck29jQSY.js";
+import { a as m, b as o, c as p } from "../../matMulGelu-BpvgnYG8.js";
+import "../../mat_mul-Bn2BDpT4.js";
 export {
   m as MATMUL_SHARED_DIM_THRESHOLD,
   o as batchMatMulGeluImpl,

package/dist/ops/webgl/matMulMul.js

@@ -1,5 +1,5 @@
-import { h as u } from "../../index-D0RBWjq8.js";
-import { b as c } from "../../matMulGelu-B2s_80-H.js";
+import { f as u } from "../../index-Duu1Lvvv.js";
+import { b as c } from "../../matMulGelu-BpvgnYG8.js";
 const M = `
     return a * b;
 `;

package/dist/ops/webgl/mulDropout.js

@@ -1,4 +1,4 @@
-import { h as m } from "../../index-D0RBWjq8.js";
+import { f as m } from "../../index-Duu1Lvvv.js";
 class f {
   variableNames = ["a", "b"];
   outputShape;

package/dist/ops/webgl/normRMS.js

@@ -1,5 +1,5 @@
-import { h as p, e as G } from "../../index-D0RBWjq8.js";
-import { s as x } from "../../sum-BdplSvq_.js";
+import { f as p, e as G } from "../../index-Duu1Lvvv.js";
+import { s as x } from "../../sum-BPUfDB2X.js";
 class y {
   variableNames = ["x", "meanSquare", "gamma"];
   outputShape;

package/dist/ops/webgl/qkv.js

@@ -1,4 +1,4 @@
-import { h as i } from "../../index-D0RBWjq8.js";
+import { f as i } from "../../index-Duu1Lvvv.js";
 class l {
   variableNames = ["x", "kernel"];
   outputShape;

package/dist/ops/webgl/rope.js

@@ -1,4 +1,4 @@
-import { h as l } from "../../index-D0RBWjq8.js";
+import { f as h } from "../../index-Duu1Lvvv.js";
 class g {
   variableNames = ["x", "sin", "cos"];
   outputShape;
@@ -45,12 +45,12 @@ class g {
   }
 }
 function f(o) {
-  const { x: t } = o.inputs, { pastLen: s, ropeCache: e, negSin: n } = o.attrs, a = n ? e.getNegSin() : e.getSin(), r = e.getCos(), d = o.backend, i = t.shape[0], c = t.shape[1], p = t.shape[2], u = t.shape[3], h = new g(i, c, p, u);
-  return d.runWebGLProgram(h, [t, a, r], "float32", [[s]]);
+  const { x: t } = o.inputs, { pastLen: s, ropeCache: e, negSin: n } = o.attrs, a = n ? e.getNegSin() : e.getSin(), r = e.getCos(), d = o.backend, i = t.shape[0], c = t.shape[1], p = t.shape[2], u = t.shape[3], l = new g(i, c, p, u);
+  return d.runWebGLProgram(l, [t, a, r], "float32", [[s]]);
 }
 const m = {
   kernelName: "Rope",
   backendName: "webgl",
   kernelFunc: f
 };
-l(m);
+h(m);

package/dist/ops/webgl/scatterSub.js

@@ -1,4 +1,4 @@
-import { h as i } from "../../index-D0RBWjq8.js";
+import { f as i } from "../../index-Duu1Lvvv.js";
 class u {
   variableNames = ["labels", "softmaxProbs", "dy"];
   outputShape;

package/dist/ops/webgpu/adamAdjust.js

@@ -1,6 +1,6 @@
-import { e as p } from "../../webgpu_program-CAE4RICo.js";
-import { f as d, c as l } from "../../webgpu_util-BdovYhXr.js";
-import { h as f, a7 as c } from "../../index-D0RBWjq8.js";
+import { e as p } from "../../webgpu_program-BpWRlghH.js";
+import { f as d, c as l } from "../../webgpu_util-DMiKzzQM.js";
+import { f, a6 as c } from "../../index-Duu1Lvvv.js";
 class h {
   variableNames = ["moments", "value"];
   outputShape;

package/dist/ops/webgpu/adamMoments.js

@@ -1,6 +1,6 @@
-import { e as u } from "../../webgpu_program-CAE4RICo.js";
-import { f as p, c as d } from "../../webgpu_util-BdovYhXr.js";
-import { h as c, a7 as f } from "../../index-D0RBWjq8.js";
+import { e as u } from "../../webgpu_program-BpWRlghH.js";
+import { f as p, c as d } from "../../webgpu_util-DMiKzzQM.js";
+import { f, a6 as c } from "../../index-Duu1Lvvv.js";
 class l {
   variableNames = ["moments", "gradient"];
   outputShape;
@@ -42,7 +42,7 @@ function h(e) {
   const { moments: t, gradient: a } = e.inputs, { beta1: n, beta2: o, lossScaling: s } = e.attrs, r = e.backend;
   if (a.dtype !== "float32")
     throw new Error(`Gradient must be float32, but got ${a.dtype}`);
-  if (f(t.shape, [...a.shape, 2], "Error in AdamMoments: "), n < 0 || n >= 1)
+  if (c(t.shape, [...a.shape, 2], "Error in AdamMoments: "), n < 0 || n >= 1)
     throw new Error(`Invalid beta1 value: ${n}. Must be in the range [0, 1).`);
   if (o < 0 || o >= 1)
     throw new Error(`Invalid beta2 value: ${o}. Must be in the range [0, 1).`);
@@ -58,4 +58,4 @@ const g = {
   backendName: "webgpu",
   kernelFunc: h
 };
-c(g);
+f(g);

package/dist/ops/webgpu/add16.js

@@ -1,4 +1,4 @@
-import { h as t } from "../../index-D0RBWjq8.js";
+import { f as t } from "../../index-Duu1Lvvv.js";
 import { BinaryOpProgram as p } from "./utils/binary_op.js";
 import { B as s } from "../../binary_op_util-pKXltfxI.js";
 function c(e) {

package/dist/ops/webgpu/appendCache.js

@@ -1,7 +1,7 @@
 import { isPackedTensor as T } from "../../utilities/packed.js";
-import { e as p } from "../../webgpu_program-CAE4RICo.js";
-import { f as d, c as u } from "../../webgpu_util-BdovYhXr.js";
-import { h as S, a7 as g } from "../../index-D0RBWjq8.js";
+import { e as p } from "../../webgpu_program-BpWRlghH.js";
+import { f as d, c as u } from "../../webgpu_util-DMiKzzQM.js";
+import { f as S, a6 as g } from "../../index-Duu1Lvvv.js";
 class x {
   variableNames = ["cache", "item"];
   outputShape;

package/dist/ops/webgpu/attentionMask.js

@@ -1,6 +1,6 @@
-import { h as d, a7 as b } from "../../index-D0RBWjq8.js";
+import { f as d, a6 as b } from "../../index-Duu1Lvvv.js";
 import { isPackedTensor as p } from "../../utilities/packed.js";
-import { b as l } from "../../matMul16-cDxwemKj.js";
+import { b as l } from "../../matMul16-xswmhSuF.js";
 import M from "./attentionMask32_program.js";
 function w(n) {
   const { q: t, k: e } = n.inputs, { divisor: a, pastLen: o } = n.attrs, m = n.backend;
@@ -11,12 +11,12 @@ function w(n) {
     throw new Error("Divisor must be non-zero in AttentionMask");
   if (o < 0)
     throw new Error("pastLen must be non-negative in AttentionMask");
-  const u = new M(s, i, k, r, c), h = [
+  const u = new M(s, i, k, r, c), f = [
     { type: "float32", data: [a] },
     { type: "int32", data: [o] },
     { type: "float32", data: [Number.NEGATIVE_INFINITY] }
-  ], f = t.dtype;
-  return m.runWebGPUProgram(u, [t, e], f, h);
+  ], h = t.dtype;
+  return m.runWebGPUProgram(u, [t, e], h, f);
 }
 const A = {
   kernelName: "AttentionMask",

package/dist/ops/webgpu/attentionMask32_program.js

@@ -1,5 +1,5 @@
-import { e as r } from "../../webgpu_program-CAE4RICo.js";
-import { f as a, c as u } from "../../webgpu_util-BdovYhXr.js";
+import { e as r } from "../../webgpu_program-BpWRlghH.js";
+import { f as a, c as u } from "../../webgpu_util-DMiKzzQM.js";
 class p {
   variableNames = ["q", "k"];
   outputShape;

package/dist/ops/webgpu/concat16.js

@@ -1,8 +1,8 @@
-import { h as x, af as I, V as c } from "../../index-D0RBWjq8.js";
-import { e as D } from "../../webgpu_program-CAE4RICo.js";
-import { f as $, c as F } from "../../webgpu_util-BdovYhXr.js";
-import { r as g } from "../../Reshape-Cd6e-Otn.js";
-import { a as L, c as d } from "../../concat_util-CHsJFZJJ.js";
+import { f as x, af as I, U as c } from "../../index-Duu1Lvvv.js";
+import { e as D } from "../../webgpu_program-BpWRlghH.js";
+import { f as $, c as F } from "../../webgpu_util-DMiKzzQM.js";
+import { r as g } from "../../Reshape-Bd4V_4X7.js";
+import { a as L, c as d } from "../../concat_util-D0je5Ppu.js";
 class T {
   outputShape;
   shaderKey;

package/dist/ops/webgpu/gatherSub.js

@@ -1,6 +1,6 @@
-import { e as u } from "../../webgpu_program-CAE4RICo.js";
-import { f as h, c as p } from "../../webgpu_util-BdovYhXr.js";
-import { h as c, a7 as r } from "../../index-D0RBWjq8.js";
+import { e as u } from "../../webgpu_program-BpWRlghH.js";
+import { f as h, c as p } from "../../webgpu_util-DMiKzzQM.js";
+import { f as c, a6 as r } from "../../index-Duu1Lvvv.js";
 class l {
   variableNames = ["labels", "logits", "values"];
   outputShape;

package/dist/ops/webgpu/gelu.js

@@ -1,6 +1,6 @@
-import { h as d } from "../../index-D0RBWjq8.js";
-import { e as s } from "../../webgpu_program-CAE4RICo.js";
-import { f as n, c as o } from "../../webgpu_util-BdovYhXr.js";
+import { f as d } from "../../index-Duu1Lvvv.js";
+import { e as s } from "../../webgpu_program-BpWRlghH.js";
+import { f as n, c as o } from "../../webgpu_util-DMiKzzQM.js";
 import { isPackedTensor as l } from "../../utilities/packed.js";
 const u = 0.7978845608028654, r = 0.044715;
 class c {

package/dist/ops/webgpu/matMul16.js

@@ -1,25 +1,25 @@
-import { h as H, m as P, b as B, V as y, $ as J } from "../../index-D0RBWjq8.js";
-import { isPackedTensor as R } from "../../utilities/packed.js";
-import { reshape16 as U } from "../reshape16.js";
-import { matMulMul as Q } from "../matMulMul.js";
+import { f as J, m as P, b as B, U, _ as Q } from "../../index-Duu1Lvvv.js";
+import { isPackedTensor as y } from "../../utilities/packed.js";
+import { reshape16 as R } from "../reshape16.js";
+import { matMulMul as V } from "../matMulMul.js";
 import { matMulGelu as X } from "../matMulGelu.js";
 import Y from "./matMul16_program.js";
-import { m as Z } from "../../mat_mul-DxpNTCRz.js";
-import { r as x } from "../../reshape-CkjKPPqB.js";
-import { t as C } from "../../transpose-JawVKyZy.js";
-function _(p) {
-  const { A: e, B: s } = p.inputs, { transposeA: d, transposeB: f, scale: i, activation: k, scaleA: c, scaleB: u, forceOutputShape: o, perm: h, causalMask: g, pastLen: E } = p.attrs, F = p.backend, S = !R(e), M = !R(s);
+import { m as Z } from "../../mat_mul-Bn2BDpT4.js";
+import { r as x } from "../../reshape-BI0yzp1T.js";
+import { t as C } from "../../transpose-BUkQCJp9.js";
+function $(p) {
+  const { A: e, B: s } = p.inputs, { transposeA: d, transposeB: f, scale: i, activation: k, scaleA: c, scaleB: u, forceOutputShape: o, perm: h, causalMask: g, pastLen: E } = p.attrs, F = p.backend, S = !y(e), M = !y(s);
   if (S && M) {
     const A = c !== void 0 ? P(e, B(c)) : e, b = u !== void 0 ? P(s, B(u)) : s;
     if (g)
       throw new Error("Causal mask is not supported for unpacked MatMul16.");
     let a;
-    if (i !== void 0 ? a = Q(A, b, B(i), d, f) : k === "gelu" ? a = X(A, b) : a = Z(A, b, d, f), h)
+    if (i !== void 0 ? a = V(A, b, B(i), d, f) : k === "gelu" ? a = X(A, b) : a = Z(A, b, d, f), h)
       if (o) {
         const r = x(a, o);
         a.dispose();
-        const q = C(r, h);
-        return r.dispose(), q;
+        const H = C(r, h);
+        return r.dispose(), H;
       } else {
         const r = C(a, h);
         return a.dispose(), r;
@@ -34,23 +34,23 @@ function _(p) {
     throw new Error("When using mixed precision, A must be packed if B is packed.");
   if (!S && M)
     throw new Error("When using mixed precision, B must be packed if A is packed.");
-  const l = e.shape.length, m = s.shape.length, W = e.shape.slice(0, -2), z = s.shape.slice(0, -2), v = y(W), w = y(z), I = J(e.shape.slice(0, -2), s.shape.slice(0, -2)), N = Math.max(v, w), K = e.shape[l - 2], L = s.shape[m - 2], T = e.shape[l - 1] * 2, V = s.shape[m - 1] * 2, D = U(e, [v, e.shape[l - 2], e.shape[l - 1]]), G = U(s, [w, s.shape[m - 2], s.shape[m - 1]]), t = new Y(N, K, L, T, V, d, f), n = [];
+  const l = e.shape.length, m = s.shape.length, W = e.shape.slice(0, -2), z = s.shape.slice(0, -2), v = U(W), w = U(z), I = Q(e.shape.slice(0, -2), s.shape.slice(0, -2)), N = Math.max(v, w), K = e.shape[l - 2], L = s.shape[m - 2], T = e.shape[l - 1] * 2, _ = s.shape[m - 1] * 2, D = R(e, [v, e.shape[l - 2], e.shape[l - 1]]), G = R(s, [w, s.shape[m - 2], s.shape[m - 1]]), t = new Y(N, K, L, T, _, d, f), n = [];
   i !== void 0 && (t.useScale(), n.push({ type: "float32", data: [i] })), c !== void 0 && (t.useScaleA(), n.push({ type: "float32", data: [c] })), u !== void 0 && (t.useScaleB(), n.push({ type: "float32", data: [u] })), k !== void 0 && t.useActivation(k), g && (t.useCausalMask(), n.push({ type: "int32", data: [E || 0] }));
   const O = t.outputShape.length;
   o && (p.attrs.originalShape = t.outputShape);
-  const $ = o ?? I.concat([t.outputShape[O - 2], t.outputShape[O - 1]]);
-  t.setOutputShape($, h);
-  const j = F.runWebGPUProgram(
+  const j = o ?? I.concat([t.outputShape[O - 2], t.outputShape[O - 1]]);
+  t.setOutputShape(j, h);
+  const q = F.runWebGPUProgram(
     t,
     [D, G],
     "packedF16",
     n.length > 0 ? n : void 0
   );
-  return D.dispose(), G.dispose(), j;
+  return D.dispose(), G.dispose(), q;
 }
 const ee = {
   kernelName: "MatMul16",
   backendName: "webgpu",
-  kernelFunc: _
+  kernelFunc: $
 };
-H(ee);
+J(ee);

package/dist/ops/webgpu/matMul16_program.js

@@ -1,5 +1,5 @@
-import { V as f } from "../../index-D0RBWjq8.js";
-import { e as h } from "../../webgpu_program-CAE4RICo.js";
+import { U as f } from "../../index-Duu1Lvvv.js";
+import { e as h } from "../../webgpu_program-BpWRlghH.js";
 class B {
   variableNames = ["A", "B"];
   outputShape;

package/dist/ops/webgpu/mul16.js

@@ -1,4 +1,4 @@
-import { h as t } from "../../index-D0RBWjq8.js";
+import { f as t } from "../../index-Duu1Lvvv.js";
 import { BinaryOpProgram as m } from "./utils/binary_op.js";
 import { B as p } from "../../binary_op_util-pKXltfxI.js";
 function s(e) {

package/dist/ops/webgpu/normRMS.js

@@ -1,7 +1,7 @@
-import { h as g, a7 as l } from "../../index-D0RBWjq8.js";
+import { f as g, a6 as l } from "../../index-Duu1Lvvv.js";
 import { createReduceInfo as w, reduce as S } from "./utils/reductions.js";
 import { isPackedTensor as d } from "../../utilities/packed.js";
-import { p as f } from "../../pack16-k4jq6aMX.js";
+import { p as f } from "../../pack16-atD0eYRm.js";
 import z from "./normRMS16_program.js";
 import N from "./normRMS32_program.js";
 import b from "./utils/deviceInfo.js";

package/dist/ops/webgpu/normRMSGrad.js

@@ -1,8 +1,8 @@
-import { h as _, a7 as y, e as D } from "../../index-D0RBWjq8.js";
+import { f as _, a6 as y, e as D } from "../../index-Duu1Lvvv.js";
 import { createReduceInfo as X } from "./utils/reductions.js";
-import { f as $ } from "../../webgpu_util-BdovYhXr.js";
-import { e as M } from "../../webgpu_program-CAE4RICo.js";
-import { p as k, u as R } from "../../pack16-k4jq6aMX.js";
+import { f as $ } from "../../webgpu_util-DMiKzzQM.js";
+import { e as M } from "../../webgpu_program-BpWRlghH.js";
+import { p as k, u as R } from "../../pack16-atD0eYRm.js";
 import { isPackedTensor as h } from "../../utilities/packed.js";
 import { reshape16 as L } from "../reshape16.js";
 import { sum16 as P } from "../sum16.js";

package/dist/ops/webgpu/pack16.js

@@ -1,4 +1,4 @@
-import { h as i } from "../../index-D0RBWjq8.js";
+import { f as i } from "../../index-Duu1Lvvv.js";
 import p from "./pack16_program.js";
 function m(e) {
   const { x: n } = e.inputs, { scaling: a, padding: r } = e.attrs, s = e.backend;
@@ -10,9 +10,9 @@ function m(e) {
   const c = [{ type: "float32", data: [a] }];
   return s.runWebGPUProgram(t, [n], "packedF16", o ? c : void 0);
 }
-const k = {
+const f = {
   kernelName: "Pack16",
   backendName: "webgpu",
   kernelFunc: m
 };
-i(k);
+i(f);

package/dist/ops/webgpu/pack16_program.js

@@ -1,5 +1,5 @@
-import { f as o, c as a } from "../../webgpu_util-BdovYhXr.js";
-import { e as s } from "../../webgpu_program-CAE4RICo.js";
+import { f as o, c as a } from "../../webgpu_util-DMiKzzQM.js";
+import { e as s } from "../../webgpu_program-BpWRlghH.js";
 class h {
   outputShape;
   shaderKey = "Pack16";