@genai-fi/nanogpt 0.10.3 → 0.12.0

package/dist/{matMulGelu-CzfgT6Wq.js → matMulGelu-BpvgnYG8.js}

@@ -1,11 +1,11 @@
-import { p as C, t as R, e as I, h as G, a3 as L, l as F, ak as U } from "./index-DOvlwCh-.js";
-import { r as M } from "./Reshape-Zt6eb7yh.js";
-import { u as H } from "./gpgpu_math-DAOmgtXR.js";
-import { m as B } from "./mat_mul-SjpJRLyL.js";
+import { f as C, t as R, e as I, U as G, _ as L, x as U, ak as F } from "./index-Duu1Lvvv.js";
+import { r as M } from "./Reshape-Ck29jQSY.js";
+import { u as H } from "./gpgpu_math-BJSTk_mW.js";
+import { m as B } from "./mat_mul-Bn2BDpT4.js";
 class W {
   constructor(e, s, a, n = !1, o = !1, r = !1, i = null, u = !1, l = !1) {
     this.variableNames = ["matrixA", "matrixB"], this.packedInputs = !0, this.packedOutput = !0, this.outputShape = a, this.enableShapeUniforms = H(this.outputShape.length);
-    const p = n ? e[1] : e[2], h = Math.ceil(p / 2), d = n ? "i * 2, rc.y" : "rc.y, i * 2", x = o ? "rc.z, i * 2" : "i * 2, rc.z", b = n ? ["a.xxyy", "a.zzww"] : ["a.xxzz", "a.yyww"], m = o ? ["b.xzxz", "b.ywyw"] : ["b.xyxy", "b.zwzw"];
+    const p = n ? e[1] : e[2], h = Math.ceil(p / 2), d = n ? "i * 2, rc.y" : "rc.y, i * 2", b = o ? "rc.z, i * 2" : "i * 2, rc.z", x = n ? ["a.xxyy", "a.zzww"] : ["a.xxzz", "a.yyww"], m = o ? ["b.xzxz", "b.ywyw"] : ["b.xyxy", "b.zwzw"];
     let c = "", g = "";
     i && (u ? c = `vec4 activation(vec4 a) {
           vec4 b = getPreluActivationWeightsAtOutCoords();
@@ -30,12 +30,12 @@ class W {
         int batchB = ${v};
         for (int i = 0; i < ${h}; i++) {
           vec4 a = getMatrixA(batchA, ${d});
-          vec4 b = getMatrixB(batchB, ${x});
+          vec4 b = getMatrixB(batchB, ${b});
 
           // These swizzled products need to be separately added.
           // See: https://github.com/tensorflow/tfjs/issues/1735
-          result += (${b[0]} * ${m[0]});
-          result += (${b[1]} * ${m[1]});
+          result += (${x[0]} * ${m[0]});
+          result += (${x[1]} * ${m[1]});
         }
         return result;
       }
@@ -90,24 +90,24 @@ function O({
   activationSnippet: o,
   multiplier: r
 }) {
-  const i = t.shape.length, u = e.shape.length, l = s ? t.shape[i - 2] : t.shape[i - 1], p = a ? e.shape[u - 1] : e.shape[u - 2], h = s ? t.shape[i - 1] : t.shape[i - 2], d = a ? e.shape[u - 2] : e.shape[u - 1], x = t.shape.slice(0, -2), b = e.shape.slice(0, -2), m = G(x), c = G(b), $ = L(t.shape.slice(0, -2), e.shape.slice(0, -2)).concat([h, d]);
-  F(
+  const i = t.shape.length, u = e.shape.length, l = s ? t.shape[i - 2] : t.shape[i - 1], p = a ? e.shape[u - 1] : e.shape[u - 2], h = s ? t.shape[i - 1] : t.shape[i - 2], d = a ? e.shape[u - 2] : e.shape[u - 1], b = t.shape.slice(0, -2), x = e.shape.slice(0, -2), m = G(b), c = G(x), $ = L(t.shape.slice(0, -2), e.shape.slice(0, -2)).concat([h, d]);
+  U(
     l === p,
     () => `Error in matMul: inner shapes (${l}) and (${p}) of Tensors with shapes ${t.shape} and ${e.shape} and transposeA=${s} and transposeB=${a} must match.`
   );
-  const f = s ? [m, l, h] : [m, h, l], v = a ? [c, d, p] : [c, p, d], A = M({ inputs: { x: t }, backend: n, attrs: { shape: f } }), y = M({ inputs: { x: e }, backend: n, attrs: { shape: v } }), k = [A, y], E = Math.max(m, c), N = o, T = U(t.dtype, e.dtype), _ = new W(
+  const f = s ? [m, l, h] : [m, h, l], v = a ? [c, d, p] : [c, p, d], A = M({ inputs: { x: t }, backend: n, attrs: { shape: f } }), y = M({ inputs: { x: e }, backend: n, attrs: { shape: v } }), k = [A, y], _ = Math.max(m, c), E = o, N = F(t.dtype, e.dtype), T = new W(
     f,
     v,
-    [E, h, d],
+    [_, h, d],
     s,
     a,
     !1,
-    N,
+    E,
     !!r,
     !1
   ), D = [A, y];
   r && D.push(r);
-  const z = n.runWebGLProgram(_, D, T), K = M({ inputs: { x: z }, backend: n, attrs: { shape: $ } });
+  const z = n.runWebGLProgram(T, D, N), K = M({ inputs: { x: z }, backend: n, attrs: { shape: $ } });
   k.push(z);
   for (const P of k)
     n.disposeIntermediateTensorInfo(P);

package/dist/mat_mul-Bn2BDpT4.js

@@ -0,0 +1,11 @@
+import { o as m, q as s, B as c, E as M, D as p } from "./index-Duu1Lvvv.js";
+function f(e, o, n = !1, l = !1) {
+  let a = s(e, "a", "matMul"), t = s(o, "b", "matMul");
+  [a, t] = c(a, t);
+  const r = { a, b: t }, u = { transposeA: n, transposeB: l };
+  return M.runKernel(p, r, u);
+}
+const i = /* @__PURE__ */ m({ matMul_: f });
+export {
+  i as m
+};

package/dist/{mod-AnXEvvpo.js → mod-B4AUd1Np.js}

@@ -1,4 +1,4 @@
-import { A as r, B as s, L as m, E as c, N as d } from "./index-DOvlwCh-.js";
+import { o as r, q as s, B as m, E as c, M as d } from "./index-Duu1Lvvv.js";
 function p(t, e) {
   let o = s(t, "a", "mod"), a = s(e, "b", "mod");
   [o, a] = m(o, a);

package/dist/models/NanoGPTV1.js

@@ -3,11 +3,11 @@ import b from "../layers/TransformerBlock.js";
 import k from "../layers/TiedEmbedding.js";
 import w from "../layers/RoPECache.js";
 import E from "../layers/RMSNorm.js";
-import { t as l, k as u } from "../index-DOvlwCh-.js";
+import { t as l, k as u } from "../index-Duu1Lvvv.js";
 import C from "./model.js";
 import P from "../layers/PositionEmbedding.js";
 import { packingSupported as _ } from "../utilities/packed.js";
-import { p as y, u as M } from "../pack16-nQ6JaLo-.js";
+import { p as y, u as M } from "../pack16-atD0eYRm.js";
 class I extends C {
   wte;
   // Token embeddings

package/dist/models/model.js

@@ -1,23 +1,23 @@
 import m from "../layers/BaseLayer.js";
-import "../index-DOvlwCh-.js";
-import "../random_width-or-CEftb.js";
-import "../zeros_like-DvE73F4e.js";
+import "../index-Duu1Lvvv.js";
+import "../random_width-BN4wGJaW.js";
+import "../zeros_like-De4n1C3m.js";
 import "../Generator.js";
 import "../index-Cp39cXWe.js";
-import "../dataset-BcwmTGYc.js";
+import "../dataset-CVIJu7Xa.js";
 import "../ops/cpu/attentionMask.js";
 import "../ops/webgl/attentionMask.js";
 import "../ops/grads/attentionMask.js";
 import "../ops/cpu/rope.js";
 import "../ops/webgl/rope.js";
-import "../rope-Ir4mTyD1.js";
+import "../rope-DJ7Y7c-u.js";
 import "../ops/cpu/appendCache.js";
 import "../ops/webgl/appendCache.js";
 import "../ops/grads/softmax16.js";
-import "../matMul16-BWRSOCWB.js";
+import "../matMul16-xswmhSuF.js";
 import "../ops/webgl/matMul16.js";
 import "../ops/cpu/matMul16.js";
-import "../pack16-nQ6JaLo-.js";
+import "../pack16-atD0eYRm.js";
 import "../ops/transpose16.js";
 import "../ops/reshape16.js";
 import "../ops/cpu/qkv.js";
@@ -40,11 +40,11 @@ import "../ops/webgl/scatterSub.js";
 import "../ops/cpu/gatherSub.js";
 import "../ops/webgl/gatherSub.js";
 import "../ops/cpu/matMulGelu.js";
-import "../matMulGelu-CzfgT6Wq.js";
+import "../matMulGelu-BpvgnYG8.js";
 import "../ops/grads/matMulGelu.js";
 import "../ops/cpu/gelu.js";
 import "../ops/webgl/gelu.js";
-import "../gelu-CjNPL4OH.js";
+import "../gelu-BK1k-n1i.js";
 import "../ops/webgl/log.js";
 import "../checks/normRMS.js";
 import "../checks/normRMSGrad.js";

package/dist/{ones-D2rT0xk2.js → ones-CBI1AQjb.js}

@@ -1,6 +1,6 @@
-import { C as n, _ as t, h as m, E as i } from "./index-DOvlwCh-.js";
-import { c as f } from "./complex-DjxcVmoX.js";
-import { z as c } from "./zeros-KnWaWf-X.js";
+import { u as n, V as t, U as m, E as i } from "./index-Duu1Lvvv.js";
+import { c as f } from "./complex-Cyg-eQeZ.js";
+import { z as c } from "./zeros-5YROwwUH.js";
 function l(o, r = "float32") {
   if (n(o), r === "complex64") {
     const s = l(o, "float32"), a = c(o, "float32");

package/dist/ops/adamAdjust.js

@@ -1,4 +1,4 @@
-import { e as i } from "../index-DOvlwCh-.js";
+import { e as i } from "../index-Duu1Lvvv.js";
 import "./cpu/adamAdjust.js";
 import "./webgl/adamAdjust.js";
 function p(r, t, e, n, m, o) {

package/dist/ops/adamMoments.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-DOvlwCh-.js";
+import { e as t } from "../index-Duu1Lvvv.js";
 import "./cpu/adamMoments.js";
 import "./webgl/adamMoments.js";
 function s(e, n, r, m, o) {

package/dist/ops/add16.js

@@ -1,4 +1,4 @@
-import { x as t, e as o } from "../index-DOvlwCh-.js";
+import { l as t, e as o } from "../index-Duu1Lvvv.js";
 import { isPackedTensor as n } from "../utilities/packed.js";
 import "./grads/add16.js";
 function m(r, e) {

package/dist/ops/appendCache.js

@@ -1,9 +1,9 @@
-import { e as a } from "../index-DOvlwCh-.js";
+import { e as a } from "../index-Duu1Lvvv.js";
 import "./cpu/appendCache.js";
 import "./webgl/appendCache.js";
 import { isPackedTensor as c } from "../utilities/packed.js";
-import { c as t } from "../concat-BV8bt5H-.js";
-import { z as f } from "../zeros-KnWaWf-X.js";
+import { c as t } from "../concat-CSm2rMwe.js";
+import { z as f } from "../zeros-5YROwwUH.js";
 function C(r, o, n, p) {
   if (!p) {
     const e = r.shape[2], s = c(r);

package/dist/ops/attentionMask.js

@@ -1,4 +1,4 @@
-import { e as r } from "../index-DOvlwCh-.js";
+import { e as r } from "../index-Duu1Lvvv.js";
 import "./cpu/attentionMask.js";
 import "./webgl/attentionMask.js";
 import "./grads/attentionMask.js";

package/dist/ops/concat16.js

@@ -1,6 +1,6 @@
 import { isPackedTensor as o } from "../utilities/packed.js";
-import { e } from "../index-DOvlwCh-.js";
-import { c } from "../concat-BV8bt5H-.js";
+import { e } from "../index-Duu1Lvvv.js";
+import { c } from "../concat-CSm2rMwe.js";
 function p(r, n) {
   return o(r[0]) ? e().runKernel("Concat16", r, { axis: n ?? -1 }) : c(r, n);
 }

package/dist/ops/cpu/adamAdjust.js

@@ -1,18 +1,18 @@
-import { p as k, w as t, x as i, m as w, y as z } from "../../index-DOvlwCh-.js";
-function A(c) {
-  const { moments: s, value: r } = c.inputs, { beta1: l, beta2: m, epsilon: u, learningRate: d } = c.attrs, e = s.shape.length, a = new Array(e).fill(0), n = s.shape.slice();
-  n[e - 1] = 1;
+import { f as k, j as t, l as i, m as z, n as A } from "../../index-Duu1Lvvv.js";
+function C(c) {
+  const { moments: n, value: r } = c.inputs, { beta1: l, beta2: m, epsilon: u, learningRate: d } = c.attrs, e = n.shape.length, a = new Array(e).fill(0), s = n.shape.slice();
+  s[e - 1] = 1;
   const o = a.slice();
   o[e - 1] = 1;
-  const p = n.slice(), b = s.slice(a, n).squeeze([e - 1]), M = s.slice(o, p).squeeze([e - 1]), g = t(b, l), f = t(M, m);
+  const b = s.slice(), p = n.slice(a, s).squeeze([e - 1]), M = n.slice(o, b).squeeze([e - 1]), f = t(p, l), g = t(M, m);
   return i(
-    w(t(g, i(z(f), u ?? 1e-8)), -d),
+    z(t(f, i(A(g), u ?? 1e-8)), -d),
     r
   );
 }
-const C = {
+const h = {
   kernelName: "AdamAdjust",
   backendName: "cpu",
-  kernelFunc: A
+  kernelFunc: C
 };
-k(C);
+k(h);

package/dist/ops/cpu/adamMoments.js

@@ -1,16 +1,16 @@
-import { p } from "../../index-DOvlwCh-.js";
-import { s as b } from "../../stack-Cf4n9h0N.js";
-function f(t) {
+import { f as p } from "../../index-Duu1Lvvv.js";
+import { s as f } from "../../stack-CDWShFHF.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;
   const i = a.slice();
   i[e - 1] = 1;
   const r = s.slice(), l = n.slice(a, s).squeeze([e - 1]), u = n.slice(i, r).squeeze([e - 1]), M = l.mul(o).add(c.mul(1 - o)), d = u.mul(m).add(c.square().mul(1 - m));
-  return b([M, d], -1);
+  return f([M, d], -1);
 }
 const g = {
   kernelName: "AdamMoments",
   backendName: "cpu",
-  kernelFunc: f
+  kernelFunc: b
 };
 p(g);

package/dist/ops/cpu/appendCache.js

@@ -1,13 +1,13 @@
-import { p as d } from "../../index-DOvlwCh-.js";
-import { c as h } from "../../concat-BV8bt5H-.js";
+import { f as d } from "../../index-Duu1Lvvv.js";
+import { c as h } from "../../concat-CSm2rMwe.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) {
-    const f = n.slice([0, 0, 0, 0], [t, o, c, a]), m = n.slice([0, 0, c + e, 0], [t, o, i - c - e, a]), r = e < e ? s.slice([0, 0, 0, 0], [t, o, e, a]) : s, k = h([f, r, m], 2);
-    return f.dispose(), m.dispose(), r !== s && r.dispose(), k;
+    const l = n.slice([0, 0, 0, 0], [t, o, c, a]), m = n.slice([0, 0, c + e, 0], [t, o, i - c - e, a]), r = e < e ? s.slice([0, 0, 0, 0], [t, o, e, a]) : s, k = h([l, r, m], 2);
+    return l.dispose(), m.dispose(), r !== s && r.dispose(), k;
   }
-  const l = n.slice([0, 0, e, 0], [t, o, i - e, a]), C = h([l, s], 2);
-  return l.dispose(), C;
+  const f = n.slice([0, 0, e, 0], [t, o, i - e, a]), C = h([f, s], 2);
+  return f.dispose(), C;
 }
 const w = {
   kernelName: "AppendCache",

package/dist/ops/cpu/attentionMask.js

@@ -1,22 +1,22 @@
-import { p as o, q as d, b as u } from "../../index-DOvlwCh-.js";
-import { l as N } from "../../ops-B5yanEdW.js";
-import { o as b } from "../../ones-D2rT0xk2.js";
-import { z as A } from "../../zeros-KnWaWf-X.js";
-import { w as I } from "../../resize_nearest_neighbor-B19mCEg2.js";
-import { m as g } from "../../mat_mul-SjpJRLyL.js";
+import { f as o, h as d, b as u } from "../../index-Duu1Lvvv.js";
+import { l as N } from "../../ops-C2_OXuZ4.js";
+import { o as b } from "../../ones-CBI1AQjb.js";
+import { z as A } from "../../zeros-5YROwwUH.js";
+import { w as I } from "../../resize_nearest_neighbor-BA_BX-ub.js";
+import { m as g } from "../../mat_mul-Bn2BDpT4.js";
 function a(n) {
   const { q: s, k: e } = n.inputs, { divisor: r } = n.attrs, c = s.shape[2], t = e.shape[2], m = N.bandPart(b([t, t]), -1, 0).cast("bool"), i = A([t, t]), l = d([t, t], Number.NEGATIVE_INFINITY), f = I(m, i, l), k = g(s, e, !1, !0).mul(u(r)), p = f.slice([0, 0], [c, t]).expandDims(0).expandDims(0);
   return k.add(p);
 }
-const w = {
+const h = {
   kernelName: "AttentionMask",
   backendName: "cpu",
   kernelFunc: a
 };
-o(w);
-const M = {
+o(h);
+const w = {
   kernelName: "AttentionMask",
   backendName: "tensorflow",
   kernelFunc: a
 };
-o(M);
+o(w);

package/dist/ops/cpu/fusedSoftmax.js

@@ -1,17 +1,17 @@
-import { p as e } from "../../index-DOvlwCh-.js";
-import { s as m } from "../../softmax-CA5jFsLR.js";
+import { f as e } from "../../index-Duu1Lvvv.js";
+import { s as f } from "../../softmax-DfuYyjMh.js";
 function n(t) {
   const { inputs: s, attrs: a } = t, { logits: o } = s, { dim: i, dropoutRate: r } = a;
   if (!o)
     throw new Error("Error in softmax: input logits is null");
-  return r !== void 0 && r > 0 && console.warn("Dropout in fusedSoftmax not implemented for CPU backend, skipping dropout."), m(o, i);
+  return r !== void 0 && r > 0 && console.warn("Dropout in fusedSoftmax not implemented for CPU backend, skipping dropout."), f(o, i);
 }
-const f = {
+const m = {
   kernelName: "FusedSoftmax",
   backendName: "cpu",
   kernelFunc: n
 };
-e(f);
+e(m);
 const u = {
   kernelName: "FusedSoftmax",
   backendName: "tensorflow",

package/dist/ops/cpu/gatherSub.js

@@ -1,18 +1,18 @@
-import { A as u, B as c, E as g, aj as p, p as h, c as m } from "../../index-DOvlwCh-.js";
-import { r as l } from "../../range-BklejeeW.js";
-import { s as N } from "../../stack-Cf4n9h0N.js";
-function f(e, t) {
+import { o as u, q as c, E as g, aj as h, f as m, c as p } from "../../index-Duu1Lvvv.js";
+import { r as f } from "../../range-DKmP1-OQ.js";
+import { s as l } from "../../stack-CDWShFHF.js";
+function N(e, t) {
   const n = c(t, "indices", "gatherND", "int32"), s = { params: c(e, "x", "gatherND", "string_or_numeric"), indices: n };
-  return g.runKernel(p, s);
+  return g.runKernel(h, s);
 }
-const b = /* @__PURE__ */ u({ gatherND_: f });
+const b = /* @__PURE__ */ u({ gatherND_: N });
 function d(e) {
-  const { values: t, labels: n, logits: r } = e.inputs, s = n.shape[0], a = l(0, s, 1, "int32"), i = N([a, n], 1), o = b(r, i);
-  return m(t, o);
+  const { values: t, labels: n, logits: r } = e.inputs, s = n.shape[0], a = f(0, s, 1, "int32"), o = l([a, n], 1), i = b(r, o);
+  return p(t, i);
 }
 const k = {
   kernelName: "EfficientGatherSub",
   backendName: "cpu",
   kernelFunc: d
 };
-h(k);
+m(k);

package/dist/ops/cpu/gelu.js

@@ -1,4 +1,4 @@
-import { p as t, t as d } from "../../index-DOvlwCh-.js";
+import { f as t, t as d } from "../../index-Duu1Lvvv.js";
 const o = 0.7978845608028654, c = 0.044715;
 function m(r) {
   const { inputs: u } = r, { x: n } = u, e = n;
@@ -7,12 +7,12 @@ function m(r) {
     return e.mul(s);
   });
 }
-const N = {
+const p = {
   kernelName: "Gelu",
   backendName: "cpu",
   kernelFunc: m
 };
-t(N);
+t(p);
 const K = {
   kernelName: "Gelu",
   backendName: "tensorflow",
@@ -22,8 +22,8 @@ t(K);
 function i(r) {
   const { dy: u, x: n } = r.inputs;
   return d(() => {
-    const e = n.square(), l = e.mul(n), a = n.add(l.mul(c)).mul(o).tanh(), f = a.square().neg().add(1), k = e.mul(3 * c).add(1), g = a.add(1).mul(0.5), G = n.mul(f).mul(o).mul(k).mul(0.5), p = g.add(G);
-    return u.mul(p);
+    const e = n.square(), l = e.mul(n), a = n.add(l.mul(c)).mul(o).tanh(), f = a.square().neg().add(1), k = e.mul(3 * c).add(1), g = a.add(1).mul(0.5), G = n.mul(f).mul(o).mul(k).mul(0.5), N = g.add(G);
+    return u.mul(N);
   });
 }
 const x = {

package/dist/ops/cpu/matMul16.js

@@ -1,6 +1,6 @@
 import { isPackedTensor as t } from "../../utilities/packed.js";
-import { p } from "../../index-DOvlwCh-.js";
-import { m as l } from "../../mat_mul-SjpJRLyL.js";
+import { f as p } from "../../index-Duu1Lvvv.js";
+import { m as l } from "../../mat_mul-Bn2BDpT4.js";
 function m(e) {
   const { A: n, B: r } = e.inputs, { transposeA: o, transposeB: s } = e.attrs, a = !t(n), c = !t(r);
   if (a && c)

package/dist/ops/cpu/matMulGelu.js

@@ -1,6 +1,6 @@
-import { p as e, t as m } from "../../index-DOvlwCh-.js";
-import { g as M, d as i } from "../../gelu-CjNPL4OH.js";
-import { m as k } from "../../mat_mul-SjpJRLyL.js";
+import { f as e, t as m } from "../../index-Duu1Lvvv.js";
+import { g as M, d as i } from "../../gelu-BK1k-n1i.js";
+import { m as k } from "../../mat_mul-Bn2BDpT4.js";
 function c(t) {
   const { inputs: u } = t, { x: n, kernel: r } = u, a = n, l = r;
   return m(() => {

package/dist/ops/cpu/matMulMul.js

@@ -1,20 +1,20 @@
-import { p as e, t as i } from "../../index-DOvlwCh-.js";
+import { f as e, t as i } from "../../index-Duu1Lvvv.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));
 }
-const p = {
+const f = {
   kernelName: "MatMulMul",
   backendName: "cpu",
   kernelFunc: n
 };
-e(p);
-const f = {
+e(f);
+const p = {
   kernelName: "MatMulMul",
   backendName: "tensorflow",
   kernelFunc: n
 };
-e(f);
+e(p);
 const g = {
   kernelName: "MatMulMul",
   backendName: "webgpu",

package/dist/ops/cpu/mulDropout.js

@@ -1,4 +1,4 @@
-import { p as e, m as t } from "../../index-DOvlwCh-.js";
+import { f as e, m as t } from "../../index-Duu1Lvvv.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 { p as o, t as d } from "../../index-DOvlwCh-.js";
+import { f as o, t as d } from "../../index-Duu1Lvvv.js";
 function i(t) {
   const { inputs: e } = t, { x: n, gamma: s } = e, r = n, a = s;
   return d(() => {
@@ -6,12 +6,12 @@ function i(t) {
     return r.mul(u).mul(a);
   });
 }
-const f = {
+const k = {
   kernelName: "RMSNorm",
   backendName: "cpu",
   kernelFunc: i
 };
-o(f);
+o(k);
 const g = {
   kernelName: "RMSNorm",
   backendName: "tensorflow",
@@ -21,8 +21,8 @@ o(g);
 function N(t) {
   const { dy: e, x: n, gamma: s } = t.inputs;
   return d(() => {
-    const r = n.shape[n.shape.length - 1], a = n.square().mean(-1, !0), m = a.add(1e-8).rsqrt(), u = n.mul(m), l = e.mul(u).sum([0, 1]), c = e.mul(s), k = c.mul(n).sum(-1, !0).div(r);
-    return [c.mul(m).sub(n.mul(k).mul(m).div(a.add(1e-8))), l];
+    const r = n.shape[n.shape.length - 1], a = n.square().mean(-1, !0), m = a.add(1e-8).rsqrt(), u = n.mul(m), l = e.mul(u).sum([0, 1]), c = e.mul(s), f = c.mul(n).sum(-1, !0).div(r);
+    return [c.mul(m).sub(n.mul(f).mul(m).div(a.add(1e-8))), l];
   });
 }
 const S = {
@@ -31,9 +31,9 @@ const S = {
   kernelFunc: N
 };
 o(S);
-const p = {
+const R = {
   kernelName: "RMSNormGrad",
   backendName: "tensorflow",
   kernelFunc: N
 };
-o(p);
+o(R);

package/dist/ops/cpu/qkv.js

@@ -1,6 +1,6 @@
-import { p as q } from "../../index-DOvlwCh-.js";
-import { r as o } from "../../reshape-ByE68wS9.js";
-import { s as x } from "../../split-CVLc0w--.js";
+import { f as q } from "../../index-Duu1Lvvv.js";
+import { r as o } from "../../reshape-BI0yzp1T.js";
+import { s as x } from "../../split-BYrLboMq.js";
 function v(p) {
   const { x: c, kernel: K } = p.inputs, { heads: n, packed: C } = p.attrs;
   if (C)

package/dist/ops/cpu/rope.js

@@ -1,8 +1,8 @@
-import { p as I } from "../../index-DOvlwCh-.js";
-import { r as y } from "../../range-BklejeeW.js";
-import { g as F } from "../../gather-D3JcZUaI.js";
-import { s as E } from "../../stack-Cf4n9h0N.js";
-import { c as T } from "../../concat-BV8bt5H-.js";
+import { f as I } from "../../index-Duu1Lvvv.js";
+import { r as y } from "../../range-DKmP1-OQ.js";
+import { g as F } from "../../gather-BSULDalH.js";
+import { s as E } from "../../stack-CDWShFHF.js";
+import { c as T } from "../../concat-CSm2rMwe.js";
 function U(c, r, p, e, n) {
   const t = e.shape[3], s = p;
   if (s > t) return e;

package/dist/ops/cpu/scatterSub.js

@@ -1,8 +1,8 @@
-import { A as f, C as g, B as r, E as l, ai as N, p as b, c as S, m as h } from "../../index-DOvlwCh-.js";
-import { v as D } from "../../scatter_nd_util-lvSiX8q4.js";
-import { r as k } from "../../range-BklejeeW.js";
-import { s as v } from "../../stack-Cf4n9h0N.js";
-import { o as E } from "../../ones-D2rT0xk2.js";
+import { o as f, u as g, q as r, E as l, ai as N, f as b, c as S, m as h } from "../../index-Duu1Lvvv.js";
+import { v as D } from "../../scatter_nd_util-k9MUVUkn.js";
+import { r as k } from "../../range-DKmP1-OQ.js";
+import { s as v } from "../../stack-CDWShFHF.js";
+import { o as E } from "../../ones-CBI1AQjb.js";
 function I(a, e, s) {
   g(s);
   const n = r(a, "indices", "scatterND", "int32"), t = r(e, "updates", "scatterND");
@@ -10,14 +10,14 @@ function I(a, e, 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]);
+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/dot16.js

@@ -1,8 +1,8 @@
-import { b as d } from "../matMul16-BWRSOCWB.js";
+import { b as d } from "../matMul16-xswmhSuF.js";
 import { transpose16 as w } from "./transpose16.js";
 import { reshape16 as n } from "./reshape16.js";
 import { isPackedTensor as p } from "../utilities/packed.js";
-import { d as x } from "../tfjs_backend-y1cvNhLA.js";
+import { d as x } from "../tfjs_backend-806hyYve.js";
 function E(e, s, h = !1, c = !1) {
   if (!p(e) && !p(s))
     return x(e, s);

package/dist/ops/gatherSub.js

@@ -1,4 +1,4 @@
-import { e as n } from "../index-DOvlwCh-.js";
+import { e as n } from "../index-Duu1Lvvv.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-DOvlwCh-.js";
+import "../index-Duu1Lvvv.js";
 import "./cpu/gelu.js";
 import "./webgl/gelu.js";
-import { d as e, g as i } from "../gelu-CjNPL4OH.js";
+import { d as e, g as i } from "../gelu-BK1k-n1i.js";
 export {
   e as dGelu,
   i as gelu

package/dist/ops/grads/add16.js

@@ -1,11 +1,11 @@
-import { u as i, a3 as h, a4 as d } from "../../index-DOvlwCh-.js";
+import { i as u, _ as h, $ as d } from "../../index-Duu1Lvvv.js";
 import { sum16 as c } from "../sum16.js";
 import { reshape16 as p } from "../reshape16.js";
 const A = {
   kernelName: "Add16",
   inputsToSave: ["a", "b"],
-  gradFunc: (s, u) => {
-    const [t, a] = u, n = h(t.shape, a.shape);
+  gradFunc: (s, i) => {
+    const [t, a] = i, n = h(t.shape, a.shape);
     if (Array.isArray(s))
       throw new Error("Add16 gradFunc expected dy to be a Tensor but got an array");
     return { a: () => {
@@ -23,4 +23,4 @@ const A = {
     } };
   }
 };
-i(A);
+u(A);

package/dist/ops/grads/attentionMask.js

@@ -1,5 +1,5 @@
-import { u as m } from "../../index-DOvlwCh-.js";
-import { m as o } from "../../matMul16-BWRSOCWB.js";
+import { i as m } from "../../index-Duu1Lvvv.js";
+import { m as o } from "../../matMul16-xswmhSuF.js";
 import { transpose16 as c } from "../transpose16.js";
 const l = {
   kernelName: "AttentionMask",