@genai-fi/nanogpt 0.10.2 → 0.10.3

package/dist/loader/oldZipLoad.js

@@ -1,22 +1,22 @@
-import "../utilities/packed.js";
-import { H as y } from "../index-D6Q1lPZO.js";
+import { ae as y } from "../index-DOvlwCh-.js";
+import "../random_width-or-CEftb.js";
+import "../zeros_like-DvE73F4e.js";
+import "../Generator.js";
+import "../index-Cp39cXWe.js";
+import "../dataset-BcwmTGYc.js";
 import "../ops/cpu/attentionMask.js";
 import "../ops/webgl/attentionMask.js";
 import "../ops/grads/attentionMask.js";
-import "../random_width-BVV9HveY.js";
-import "../register_all_kernels-nvj2k7OC.js";
-import "../index-Cp39cXWe.js";
-import "../dataset-D2P7rHAw.js";
 import "../ops/cpu/rope.js";
 import "../ops/webgl/rope.js";
-import "../rope-s4W2XO9B.js";
+import "../rope-Ir4mTyD1.js";
 import "../ops/cpu/appendCache.js";
 import "../ops/webgl/appendCache.js";
 import "../ops/grads/softmax16.js";
-import "../matMul16-fEAJ4smh.js";
+import "../matMul16-BWRSOCWB.js";
 import "../ops/webgl/matMul16.js";
 import "../ops/cpu/matMul16.js";
-import "../pack16-CmVZs6af.js";
+import "../pack16-nQ6JaLo-.js";
 import "../ops/transpose16.js";
 import "../ops/reshape16.js";
 import "../ops/cpu/qkv.js";
@@ -35,7 +35,6 @@ import h from "../tokeniser/CharTokeniser.js";
 import k from "../tokeniser/bpe.js";
 import { dummyPassAsync as g } from "../utilities/dummy.js";
 import b from "../models/factory.js";
-import "../Generator.js";
 import "../index-DvYrXKkX.js";
 import "../ops/cpu/adamAdjust.js";
 import "../ops/webgl/adamAdjust.js";
@@ -43,16 +42,16 @@ import "../ops/cpu/adamMoments.js";
 import "../ops/webgl/adamMoments.js";
 import "../papaparse.min-C0cScC2i.js";
 import "../ops/cpu/matMulGelu.js";
-import "../ops/webgl/matMulGelu.js";
+import "../matMulGelu-CzfgT6Wq.js";
 import "../ops/grads/matMulGelu.js";
 import "../ops/cpu/gelu.js";
 import "../ops/webgl/gelu.js";
-import "../gelu-Bmhopi0J.js";
+import "../gelu-CjNPL4OH.js";
 import "../ops/webgl/log.js";
 import "../checks/normRMS.js";
 import "../checks/normRMSGrad.js";
 import { importWeights as u } from "../utilities/weights.js";
-async function Ot(r) {
+async function Mt(r) {
   const e = /* @__PURE__ */ new Map(), p = await r.file("manifest.json")?.async("string");
   if (!p)
     throw new Error("Manifest file not found in the zip archive");
@@ -75,5 +74,5 @@ async function Ot(r) {
   return await g(m), m.loadWeights(c), { model: m, tokeniser: f };
 }
 export {
-  Ot as default
+  Mt as default
 };

package/dist/log_sum_exp-ngO0-4pK.js

@@ -0,0 +1,39 @@
+import { A as r, B as x, E as p, am as E, an as $, ao as d, af as h, c as S, x as K } from "./index-DOvlwCh-.js";
+import { e as _ } from "./axis_util-BaG7mf5A.js";
+import { r as m } from "./reshape-ByE68wS9.js";
+import { s as T } from "./sum-DWAtNGez.js";
+function b(s, o = null, n = !1) {
+  const a = { x: x(s, "x", "max") }, e = { reductionIndices: o, keepDims: n };
+  return p.runKernel(E, a, e);
+}
+const A = /* @__PURE__ */ r({ max_: b });
+function I(s) {
+  const n = { x: x(s, "x", "exp") };
+  return p.runKernel($, n);
+}
+const M = /* @__PURE__ */ r({ exp_: I });
+function N(s) {
+  const n = { x: x(s, "x", "log", "float32") };
+  return p.runKernel(d, n);
+}
+const v = /* @__PURE__ */ r({ log_: N });
+function w(s, o = null, n = !1) {
+  const t = x(s, "x", "logSumExp"), a = h(o, t.shape), e = A(
+    t,
+    a,
+    !0
+    /* keepDims */
+  ), l = S(t, e), i = M(l), f = T(i, a), u = v(f), c = K(m(e, u.shape), u);
+  if (n) {
+    const g = _(c.shape, a);
+    return m(c, g);
+  }
+  return c;
+}
+const P = /* @__PURE__ */ r({ logSumExp_: w });
+export {
+  v as a,
+  M as e,
+  P as l,
+  A as m
+};

package/dist/main.js

@@ -1,15 +1,18 @@
-import "./utilities/packed.js";
-import { default as ro } from "./models/NanoGPTV1.js";
-import { default as eo } from "./TeachableLLM.js";
-import { default as po } from "./tokeniser/CharTokeniser.js";
-import { default as ao } from "./tokeniser/bpe.js";
-import { default as fo } from "./utilities/waitForModel.js";
-import { default as no } from "./data/textLoader.js";
-import { default as uo } from "./Generator.js";
-import { default as To } from "./models/model.js";
-import { estimateMemoryUsage as go, estimateParameterCount as Mo, estimateResources as Po, estimateTrainingMemoryUsage as Co, validateConfig as Eo } from "./utilities/parameters.js";
-import { default as Bo } from "./utilities/topP.js";
-import "./index-D6Q1lPZO.js";
+import "./index-DOvlwCh-.js";
+import "./random_width-or-CEftb.js";
+import "./zeros_like-DvE73F4e.js";
+import { default as oo } from "./Generator.js";
+import "./index-Cp39cXWe.js";
+import "./dataset-BcwmTGYc.js";
+import { default as to } from "./models/NanoGPTV1.js";
+import { default as mo } from "./TeachableLLM.js";
+import { default as io } from "./tokeniser/CharTokeniser.js";
+import { default as so } from "./tokeniser/bpe.js";
+import { default as lo } from "./utilities/waitForModel.js";
+import { default as xo } from "./data/textLoader.js";
+import { default as co } from "./models/model.js";
+import { estimateMemoryUsage as ko, estimateParameterCount as go, estimateResources as Mo, estimateTrainingMemoryUsage as Po, validateConfig as Co } from "./utilities/parameters.js";
+import { default as ho } from "./utilities/topP.js";
 import "./ops/cpu/scatterSub.js";
 import "./ops/webgl/scatterSub.js";
 import "./ops/cpu/gatherSub.js";
@@ -20,21 +23,17 @@ import "./ops/grads/attentionMask.js";
 import "./ops/cpu/qkv.js";
 import "./ops/webgl/qkv.js";
 import "./ops/grads/qkv.js";
-import "./random_width-BVV9HveY.js";
-import "./register_all_kernels-nvj2k7OC.js";
-import "./index-Cp39cXWe.js";
-import "./dataset-D2P7rHAw.js";
 import "./ops/cpu/rope.js";
 import "./ops/webgl/rope.js";
-import "./rope-s4W2XO9B.js";
+import "./rope-Ir4mTyD1.js";
 import "./ops/cpu/appendCache.js";
 import "./ops/webgl/appendCache.js";
 import "./ops/cpu/matMulGelu.js";
-import "./ops/webgl/matMulGelu.js";
+import "./matMulGelu-CzfgT6Wq.js";
 import "./ops/grads/matMulGelu.js";
 import "./ops/cpu/gelu.js";
 import "./ops/webgl/gelu.js";
-import "./gelu-Bmhopi0J.js";
+import "./gelu-CjNPL4OH.js";
 import "./ops/cpu/normRMS.js";
 import "./ops/webgl/normRMS.js";
 import "./ops/grads/normRMS.js";
@@ -43,51 +42,51 @@ import "./ops/cpu/adamMoments.js";
 import "./ops/webgl/adamMoments.js";
 import "./ops/cpu/adamAdjust.js";
 import "./ops/webgl/adamAdjust.js";
-import { u as o, p as r } from "./pack16-CmVZs6af.js";
+import { u as o, p as r } from "./pack16-nQ6JaLo-.js";
 import "./ops/grads/softmax16.js";
-import "./matMul16-fEAJ4smh.js";
+import "./matMul16-BWRSOCWB.js";
 import "./ops/webgl/matMul16.js";
 import "./ops/cpu/matMul16.js";
 import "./ops/transpose16.js";
-import { selectBackend as yo } from "./backend.js";
-import { default as Ao } from "./utilities/performance.js";
+import { selectBackend as bo } from "./backend.js";
+import { default as Lo } from "./utilities/performance.js";
 import t from "./layers/CausalSelfAttention.js";
 import e from "./layers/MLP.js";
 import m from "./layers/TransformerBlock.js";
 import p from "./layers/RoPECache.js";
-import { default as Ro } from "./training/AdamExt.js";
-import { default as vo } from "./checks/index.js";
-import { sentenceEmbeddings as Do, sentenceEmbeddingsTensor as Fo } from "./utilities/sentences.js";
-const Z = {
+import { default as Go } from "./training/AdamExt.js";
+import { default as Uo } from "./checks/index.js";
+import { sentenceEmbeddings as wo, sentenceEmbeddingsTensor as Do } from "./utilities/sentences.js";
+const Y = {
   pack16: r,
   unpack16: o
-}, _ = {
+}, Z = {
   CausalSelfAttention: t,
   MLP: e,
   TransformerBlock: m,
   RoPECache: p
 };
 export {
-  Ro as AdamExt,
-  ao as BPETokeniser,
-  po as CharTokeniser,
-  uo as Generator,
-  To as Model,
-  ro as NanoGPT,
-  eo as TeachableLLM,
-  vo as checks,
-  go as estimateMemoryUsage,
-  Mo as estimateParameterCount,
-  Po as estimateResources,
-  Co as estimateTrainingMemoryUsage,
-  _ as layers,
-  no as loadTextData,
-  Z as ops,
-  Ao as performanceTest,
-  yo as selectBackend,
-  Do as sentenceEmbeddings,
-  Fo as sentenceEmbeddingsTensor,
-  Bo as topP,
-  Eo as validateConfig,
-  fo as waitForModel
+  Go as AdamExt,
+  so as BPETokeniser,
+  io as CharTokeniser,
+  oo as Generator,
+  co as Model,
+  to as NanoGPT,
+  mo as TeachableLLM,
+  Uo as checks,
+  ko as estimateMemoryUsage,
+  go as estimateParameterCount,
+  Mo as estimateResources,
+  Po as estimateTrainingMemoryUsage,
+  Z as layers,
+  xo as loadTextData,
+  Y as ops,
+  Lo as performanceTest,
+  bo as selectBackend,
+  wo as sentenceEmbeddings,
+  Do as sentenceEmbeddingsTensor,
+  ho as topP,
+  Co as validateConfig,
+  lo as waitForModel
 };

package/dist/{matMul16-fEAJ4smh.js → matMul16-BWRSOCWB.js}

@@ -1,13 +1,12 @@
-import { e as y } from "./index-D6Q1lPZO.js";
+import { u as y, e as h } from "./index-DOvlwCh-.js";
 import "./ops/webgl/matMul16.js";
 import "./ops/cpu/matMul16.js";
-import { isPackedTensor as g, packTensor as k } from "./utilities/packed.js";
-import { p as v } from "./pack16-CmVZs6af.js";
-import { d as h } from "./gelu-Bmhopi0J.js";
+import { isPackedTensor as g } from "./utilities/packed.js";
+import { p as v } from "./pack16-nQ6JaLo-.js";
+import { d as k } from "./gelu-CjNPL4OH.js";
 import { transpose16 as S } from "./ops/transpose16.js";
 import { reshape16 as w } from "./ops/reshape16.js";
-import { a as G } from "./tensor_util-DfwaWayG.js";
-const T = {
+const G = {
   kernelName: "MatMul16",
   inputsToSave: ["A", "B"],
   outputsToSave: [],
@@ -30,7 +29,7 @@ const T = {
     }
     if (p === "gelu") {
       const u = e, m = l(s, t, f, i);
-      e = h(u, m), u.dispose(), m.dispose();
+      e = k(u, m), u.dispose(), m.dispose();
     }
     if (!f && !i)
       return {
@@ -50,12 +49,12 @@ const T = {
     throw new Error("Gradient for transposeA=true and transposeB=true is not supported yet.");
   }
 };
-G(T);
+y(G);
 function l(r, o, n = !1, s = !1, t = {}) {
-  const e = g(r), f = g(o), i = e || f, a = !i || e ? r : v(r), p = !i || f ? o : v(o), c = y().runKernel("MatMul16", { A: a, B: p }, { transposeA: n, transposeB: s, ...t });
-  return i && !e && a.dispose(), i && !f && p.dispose(), i ? k(c) : c;
+  const e = g(r), f = g(o), i = e || f, a = !i || e ? r : v(r), p = !i || f ? o : v(o), c = h().runKernel("MatMul16", { A: a, B: p }, { transposeA: n, transposeB: s, ...t });
+  return i && !e && a.dispose(), i && !f && p.dispose(), c;
 }
-function j(r, o, n, s = !1, t = !1) {
+function K(r, o, n, s = !1, t = !1) {
   return l(r, o, s, t, { scale: n });
 }
 function B(r, o, n, s = !1, t = !1) {
@@ -64,14 +63,14 @@ function B(r, o, n, s = !1, t = !1) {
 function M(r, o, n, s = !1, t = !1) {
   return l(r, o, s, t, { scaleB: n });
 }
-function q(r, o, n = !1, s = !1) {
+function N(r, o, n = !1, s = !1) {
   return l(r, o, n, s, { activation: "gelu" });
 }
 export {
-  T as a,
+  G as a,
   l as b,
-  q as c,
+  N as c,
   B as d,
   M as e,
-  j as m
+  K as m
 };

package/dist/matMulGelu-CzfgT6Wq.js

@@ -0,0 +1,163 @@
+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";
+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"];
+    let c = "", g = "";
+    i && (u ? c = `vec4 activation(vec4 a) {
+          vec4 b = getPreluActivationWeightsAtOutCoords();
+          ${i}
+        }` : l ? c = `vec4 activation(vec4 a) {
+          vec4 b = getLeakyreluAlphaAtOutCoords();
+          ${i}
+        }` : c = `vec4 activation(vec4 x) {
+          ${i}
+        }`, g = "result = activation(result);");
+    const $ = r ? "result += getBiasAtOutCoords();" : "";
+    r && this.variableNames.push("bias"), u && this.variableNames.push("preluActivationWeights"), l && this.variableNames.push("leakyreluAlpha");
+    let f = "rc.x", v = "rc.x";
+    e[0] < s[0] ? f = `imod(rc.x, ${e[0]})` : s[0] < e[0] && (v = `imod(rc.x, ${s[0]})`), this.userCode = `
+      ${c}
+      // Don't use uniform for sharedDimensionPacked for performance.
+      const float sharedDimension = ${h}.0;
+
+      vec4 dot2x2ARowBCol(ivec3 rc) {
+        vec4 result = vec4(0);
+        int batchA = ${f};
+        int batchB = ${v};
+        for (int i = 0; i < ${h}; i++) {
+          vec4 a = getMatrixA(batchA, ${d});
+          vec4 b = getMatrixB(batchB, ${x});
+
+          // These swizzled products need to be separately added.
+          // See: https://github.com/tensorflow/tfjs/issues/1735
+          result += (${b[0]} * ${m[0]});
+          result += (${b[1]} * ${m[1]});
+        }
+        return result;
+      }
+
+      void main() {
+        ivec3 rc = getOutputCoords();
+        vec4 result = dot2x2ARowBCol(rc);
+
+        ${$}
+
+        ${g}
+
+        setOutput(result);
+      }
+    `;
+  }
+}
+const S = 0.7978845608028654, w = 0.044715, j = `
+    vec4 x3 = x * x * x;
+    vec4 inner = x + ${w} * x3;
+    inner = ${S} * 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  = ${S} * (a + ${w} * 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 = ${S} * (1.0 + 3.0 * ${w} * a2);
+    vec4 dgelu = 0.5 * (1.0 + t) + 0.5 * a * sech2 * du_dx;
+    return dgelu * b;
+`, se = 1e3;
+function O({
+  a: t,
+  b: e,
+  transposeA: s,
+  transposeB: a,
+  backend: n,
+  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(
+    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(
+    f,
+    v,
+    [E, h, d],
+    s,
+    a,
+    !1,
+    N,
+    !!r,
+    !1
+  ), D = [A, y];
+  r && D.push(r);
+  const z = n.runWebGLProgram(_, D, T), K = M({ inputs: { x: z }, backend: n, attrs: { shape: $ } });
+  k.push(z);
+  for (const P of k)
+    n.disposeIntermediateTensorInfo(P);
+  return K;
+}
+function J(t) {
+  const { inputs: e, backend: s } = t, { x: a, kernel: n } = e;
+  if (a === void 0 || n === void 0)
+    throw new Error("BatchMatMul requires two input tensors.");
+  return O({
+    a,
+    b: n,
+    transposeA: !1,
+    transposeB: !1,
+    backend: s,
+    activationSnippet: j
+  });
+}
+const Q = {
+  kernelName: "MatMulGelu",
+  backendName: "webgl",
+  kernelFunc: J
+};
+C(Q);
+function V(t) {
+  const { dy: e, x: s, kernel: a } = t.inputs, n = t.backend;
+  return R(() => {
+    const o = I().makeTensorFromTensorInfo(
+      O({
+        a: s,
+        b: a,
+        transposeA: !1,
+        transposeB: !1,
+        backend: n,
+        activationSnippet: q,
+        multiplier: e
+      })
+    ), r = B(o, a, !1, !0), i = B(s, o, !0, !1);
+    return [r, i];
+  });
+}
+const X = {
+  kernelName: "MatMulGeluGrad",
+  backendName: "webgl",
+  kernelFunc: V
+};
+C(X);
+export {
+  W as M,
+  se as a,
+  O as b,
+  J as c
+};

package/dist/mat_mul-SjpJRLyL.js

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

package/dist/mod-AnXEvvpo.js

@@ -0,0 +1,11 @@
+import { A as r, B as s, L as m, E as c, N as d } from "./index-DOvlwCh-.js";
+function p(t, e) {
+  let o = s(t, "a", "mod"), a = s(e, "b", "mod");
+  [o, a] = m(o, a);
+  const n = { a: o, b: a };
+  return c.runKernel(d, n);
+}
+const b = /* @__PURE__ */ r({ mod_: p });
+export {
+  b as m
+};

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-D6Q1lPZO.js";
+import { t as l, k as u } from "../index-DOvlwCh-.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-CmVZs6af.js";
+import { p as y, u as M } from "../pack16-nQ6JaLo-.js";
 class I extends C {
   wte;
   // Token embeddings

package/dist/models/model.js

@@ -1,23 +1,23 @@
 import m from "../layers/BaseLayer.js";
-import "../utilities/packed.js";
-import "../index-D6Q1lPZO.js";
+import "../index-DOvlwCh-.js";
+import "../random_width-or-CEftb.js";
+import "../zeros_like-DvE73F4e.js";
+import "../Generator.js";
+import "../index-Cp39cXWe.js";
+import "../dataset-BcwmTGYc.js";
 import "../ops/cpu/attentionMask.js";
 import "../ops/webgl/attentionMask.js";
 import "../ops/grads/attentionMask.js";
-import "../random_width-BVV9HveY.js";
-import "../register_all_kernels-nvj2k7OC.js";
-import "../index-Cp39cXWe.js";
-import "../dataset-D2P7rHAw.js";
 import "../ops/cpu/rope.js";
 import "../ops/webgl/rope.js";
-import "../rope-s4W2XO9B.js";
+import "../rope-Ir4mTyD1.js";
 import "../ops/cpu/appendCache.js";
 import "../ops/webgl/appendCache.js";
 import "../ops/grads/softmax16.js";
-import "../matMul16-fEAJ4smh.js";
+import "../matMul16-BWRSOCWB.js";
 import "../ops/webgl/matMul16.js";
 import "../ops/cpu/matMul16.js";
-import "../pack16-CmVZs6af.js";
+import "../pack16-nQ6JaLo-.js";
 import "../ops/transpose16.js";
 import "../ops/reshape16.js";
 import "../ops/cpu/qkv.js";
@@ -29,7 +29,6 @@ import "../ops/grads/normRMS.js";
 import "../ops/grads/add16.js";
 import "../jszip.min-Bz5-11Bk.js";
 import "../index-DvYrXKkX.js";
-import "../Generator.js";
 import "../ops/cpu/adamAdjust.js";
 import "../ops/webgl/adamAdjust.js";
 import "../ops/cpu/adamMoments.js";
@@ -41,16 +40,16 @@ import "../ops/webgl/scatterSub.js";
 import "../ops/cpu/gatherSub.js";
 import "../ops/webgl/gatherSub.js";
 import "../ops/cpu/matMulGelu.js";
-import "../ops/webgl/matMulGelu.js";
+import "../matMulGelu-CzfgT6Wq.js";
 import "../ops/grads/matMulGelu.js";
 import "../ops/cpu/gelu.js";
 import "../ops/webgl/gelu.js";
-import "../gelu-Bmhopi0J.js";
+import "../gelu-CjNPL4OH.js";
 import "../ops/webgl/log.js";
 import "../checks/normRMS.js";
 import "../checks/normRMSGrad.js";
 import { createSoftmaxCrossEntropyWithGrad as s } from "../training/sparseCrossEntropy.js";
-class st extends m {
+class et extends m {
   lossScaling = 128;
   trainingState = null;
   getNumParams() {
@@ -74,5 +73,5 @@ class st extends m {
   }
 }
 export {
-  st as default
+  et as default
 };

package/dist/ones-D2rT0xk2.js

@@ -0,0 +1,14 @@
+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";
+function l(o, r = "float32") {
+  if (n(o), r === "complex64") {
+    const s = l(o, "float32"), a = c(o, "float32");
+    return f(s, a);
+  }
+  const e = t(m(o), r);
+  return i.makeTensor(e, o, r);
+}
+export {
+  l as o
+};

package/dist/ops/adamAdjust.js

@@ -1,4 +1,4 @@
-import { e as i } from "../index-D6Q1lPZO.js";
+import { e as i } from "../index-DOvlwCh-.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-D6Q1lPZO.js";
+import { e as t } from "../index-DOvlwCh-.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 { n as t, e as o } from "../index-D6Q1lPZO.js";
+import { x as t, e as o } from "../index-DOvlwCh-.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-D6Q1lPZO.js";
+import { e as a } from "../index-DOvlwCh-.js";
 import "./cpu/appendCache.js";
 import "./webgl/appendCache.js";
 import { isPackedTensor as c } from "../utilities/packed.js";
-import { c as t } from "../concat-DmBLPVGC.js";
-import { z as f } from "../zeros-DBFVbpv5.js";
+import { c as t } from "../concat-BV8bt5H-.js";
+import { z as f } from "../zeros-KnWaWf-X.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-D6Q1lPZO.js";
+import { e as r } from "../index-DOvlwCh-.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-D6Q1lPZO.js";
-import { c } from "../concat-DmBLPVGC.js";
+import { e } from "../index-DOvlwCh-.js";
+import { c } from "../concat-BV8bt5H-.js";
 function p(r, n) {
   return o(r[0]) ? e().runKernel("Concat16", r, { axis: n ?? -1 }) : c(r, n);
 }