@genai-fi/nanogpt 0.5.6 → 0.6.1

package/dist/{log_sum_exp-CxfBtUaG.js → log_sum_exp-CV_5-TTu.js}

@@ -1,8 +1,8 @@
-import { o as r, j as p, E as u, a6 as E, a7 as h, a1 as S, s as $, a8 as d } from "./index-Du-bmOP8.js";
-import { e as K } from "./axis_util-BczFISHz.js";
-import { m as T } from "./max-0Xnlpv8k.js";
-import { r as m } from "./reshape-BH7eBpwq.js";
-import { s as _ } from "./sum-Cvq06317.js";
+import { q as r, w as p, E as u, a8 as E, a9 as h, p as S, s as $, a7 as d } from "./index-CamYe_M8.js";
+import { e as K } from "./axis_util-CCNL7jea.js";
+import { m as T } from "./max-JBBv7aUf.js";
+import { r as m } from "./reshape-C45vIIRU.js";
+import { s as _ } from "./sum-BpcpxNEh.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -23,7 +23,7 @@ function b(s) {
   const o = { x: p(s, "x", "exp") };
   return u.runKernel(E, o);
 }
-const N = /* @__PURE__ */ r({ exp_: b });
+const w = /* @__PURE__ */ r({ exp_: b });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -40,11 +40,11 @@ const N = /* @__PURE__ */ r({ exp_: b });
  * limitations under the License.
  * =============================================================================
  */
-function j(s) {
+function N(s) {
   const o = { x: p(s, "x", "log", "float32") };
   return u.runKernel(h, o);
 }
-const v = /* @__PURE__ */ r({ log_: j });
+const q = /* @__PURE__ */ r({ log_: N });
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -61,22 +61,22 @@ const v = /* @__PURE__ */ r({ log_: j });
  * limitations under the License.
  * =============================================================================
  */
-function w(s, a = null, o = !1) {
-  const n = p(s, "x", "logSumExp"), t = S(a, n.shape), x = T(
-    n,
+function v(s, n = null, o = !1) {
+  const a = p(s, "x", "logSumExp"), t = S(n, a.shape), x = T(
+    a,
     t,
     !0
     /* keepDims */
-  ), i = $(n, x), l = N(i), f = _(l, t), c = v(f), e = d(m(x, c.shape), c);
+  ), i = $(a, x), l = w(i), f = _(l, t), c = q(f), e = d(m(x, c.shape), c);
   if (o) {
     const g = K(e.shape, t);
     return m(e, g);
   }
   return e;
 }
-const M = /* @__PURE__ */ r({ logSumExp_: w });
+const M = /* @__PURE__ */ r({ logSumExp_: v });
 export {
-  v as a,
-  N as e,
+  q as a,
+  w as e,
   M as l
 };

package/dist/main.js

@@ -1,11 +1,11 @@
-import { default as E } from "./NanoGPTModel.js";
-import { default as G } from "./TeachableLLM.js";
-import { default as R } from "./tokeniser/CharTokeniser.js";
-import { default as q } from "./tokeniser/bpe.js";
-import { default as A } from "./utilities/waitForModel.js";
-import { default as I } from "./data/textLoader.js";
-import { estimateMemoryUsage as K, estimateParameterCount as O, estimateResources as Q, estimateTrainingMemoryUsage as S, validateConfig as V } from "./utilities/parameters.js";
-import "./index-Du-bmOP8.js";
+import { default as R } from "./NanoGPTModel.js";
+import { default as q } from "./TeachableLLM.js";
+import { default as A } from "./tokeniser/CharTokeniser.js";
+import { default as I } from "./tokeniser/bpe.js";
+import { default as K } from "./utilities/waitForModel.js";
+import { default as Q } from "./data/textLoader.js";
+import { estimateMemoryUsage as V, estimateParameterCount as W, estimateResources as X, estimateTrainingMemoryUsage as Y, validateConfig as Z } from "./utilities/parameters.js";
+import "./index-CamYe_M8.js";
 import "./ops/cpu/scatterSub.js";
 import "./ops/webgl/scatterSub.js";
 import "./ops/cpu/gatherSub.js";
@@ -16,7 +16,10 @@ import "./ops/grads/attentionMask.js";
 import "./ops/cpu/qkv.js";
 import "./ops/webgl/qkv.js";
 import "./ops/grads/qkv.js";
-import "@tensorflow/tfjs";
+import "./random_width-CLMQG5Jn.js";
+import "./register_all_kernels-COt6wLD0.js";
+import "./index-Tf7vU29b.js";
+import "./dataset-BFFipD1c.js";
 import "./ops/cpu/rope.js";
 import "./ops/webgl/rope.js";
 import "./ops/grads/rope.js";
@@ -34,16 +37,17 @@ import "./ops/grads/gelu.js";
 import "./ops/cpu/normRMS.js";
 import "./ops/webgl/normRMS.js";
 import "./ops/grads/normRMS.js";
+import "./ops/webgl/log.js";
 export {
-  q as BPETokeniser,
-  R as CharTokeniser,
-  E as NanoGPT,
-  G as TeachableLLM,
-  K as estimateMemoryUsage,
-  O as estimateParameterCount,
-  Q as estimateResources,
-  S as estimateTrainingMemoryUsage,
-  I as loadTextData,
-  V as validateConfig,
-  A as waitForModel
+  I as BPETokeniser,
+  A as CharTokeniser,
+  R as NanoGPT,
+  q as TeachableLLM,
+  V as estimateMemoryUsage,
+  W as estimateParameterCount,
+  X as estimateResources,
+  Y as estimateTrainingMemoryUsage,
+  Q as loadTextData,
+  Z as validateConfig,
+  K as waitForModel
 };

package/dist/{mat_mul-CbiqIe2d.js → mat_mul-CAbRFWUj.js}

@@ -1,4 +1,4 @@
-import { o as m, j as s, u as c, E as M, B as p } from "./index-Du-bmOP8.js";
+import { q as m, w as s, C as c, E as M, D as p } from "./index-CamYe_M8.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -15,10 +15,10 @@ import { o as m, j as s, u as c, E as M, B as p } from "./index-Du-bmOP8.js";
  * limitations under the License.
  * =============================================================================
  */
-function f(e, o, n = !1, l = !1) {
-  let a = s(e, "a", "matMul"), t = s(o, "b", "matMul");
+function f(e, n, o = !1, l = !1) {
+  let a = s(e, "a", "matMul"), t = s(n, "b", "matMul");
   [a, t] = c(a, t);
-  const r = { a, b: t }, u = { transposeA: n, transposeB: l };
+  const r = { a, b: t }, u = { transposeA: o, transposeB: l };
   return M.runKernel(p, r, u);
 }
 const i = /* @__PURE__ */ m({ matMul_: f });

package/dist/{max-0Xnlpv8k.js → max-JBBv7aUf.js}

@@ -1,4 +1,4 @@
-import { o as r, j as e, E as x, M as c } from "./index-Du-bmOP8.js";
+import { q as r, w as e, E as x, M as c } from "./index-CamYe_M8.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -15,8 +15,8 @@ import { o as r, j as e, E as x, M as c } from "./index-Du-bmOP8.js";
  * limitations under the License.
  * =============================================================================
  */
-function m(n, o = null, s = !1) {
-  const t = { x: e(n, "x", "max") }, a = { reductionIndices: o, keepDims: s };
+function m(n, s = null, o = !1) {
+  const t = { x: e(n, "x", "max") }, a = { reductionIndices: s, keepDims: o };
   return x.runKernel(c, t, a);
 }
 const l = /* @__PURE__ */ r({ max_: m });

package/dist/mulmat_packed_gpu-DW4doKL_.js

@@ -0,0 +1,71 @@
+import { u as z } from "./gpgpu_math-B_ycgZ4W.js";
+/**
+ * @license
+ * Copyright 2018 Google LLC. All Rights Reserved.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ * =============================================================================
+ */
+class g {
+  constructor(e, s, v, a = !1, r = !1, c = !1, t = null, o = !1, l = !1) {
+    this.variableNames = ["matrixA", "matrixB"], this.packedInputs = !0, this.packedOutput = !0, this.outputShape = v, this.enableShapeUniforms = z(this.outputShape.length);
+    const m = a ? e[1] : e[2], u = Math.ceil(m / 2), b = a ? "i * 2, rc.y" : "rc.y, i * 2", x = r ? "rc.z, i * 2" : "i * 2, rc.z", n = a ? ["a.xxyy", "a.zzww"] : ["a.xxzz", "a.yyww"], p = r ? ["b.xzxz", "b.ywyw"] : ["b.xyxy", "b.zwzw"];
+    let i = "", h = "";
+    t && (o ? i = `vec4 activation(vec4 a) {
+          vec4 b = getPreluActivationWeightsAtOutCoords();
+          ${t}
+        }` : l ? i = `vec4 activation(vec4 a) {
+          vec4 b = getLeakyreluAlphaAtOutCoords();
+          ${t}
+        }` : i = `vec4 activation(vec4 x) {
+          ${t}
+        }`, h = "result = activation(result);");
+    const $ = c ? "result += getBiasAtOutCoords();" : "";
+    c && this.variableNames.push("bias"), o && this.variableNames.push("preluActivationWeights"), l && this.variableNames.push("leakyreluAlpha");
+    let d = "rc.x", f = "rc.x";
+    e[0] < s[0] ? d = `imod(rc.x, ${e[0]})` : s[0] < e[0] && (f = `imod(rc.x, ${s[0]})`), this.userCode = `
+      ${i}
+      // Don't use uniform for sharedDimensionPacked for performance.
+      const float sharedDimension = ${u}.0;
+
+      vec4 dot2x2ARowBCol(ivec3 rc) {
+        vec4 result = vec4(0);
+        int batchA = ${d};
+        int batchB = ${f};
+        for (int i = 0; i < ${u}; i++) {
+          vec4 a = getMatrixA(batchA, ${b});
+          vec4 b = getMatrixB(batchB, ${x});
+
+          // These swizzled products need to be separately added.
+          // See: https://github.com/tensorflow/tfjs/issues/1735
+          result += (${n[0]} * ${p[0]});
+          result += (${n[1]} * ${p[1]});
+        }
+        return result;
+      }
+
+      void main() {
+        ivec3 rc = getOutputCoords();
+        vec4 result = dot2x2ARowBCol(rc);
+
+        ${$}
+
+        ${h}
+
+        setOutput(result);
+      }
+    `;
+  }
+}
+export {
+  g as M
+};

package/dist/{norm-01kY9I2B.js → norm-B9dQTFYn.js}

@@ -1,8 +1,8 @@
-import { o as l, j as c, E as y, a0 as E, a1 as w, a2 as o, a3 as u, W as v, f as I, a4 as A } from "./index-Du-bmOP8.js";
-import { e as $ } from "./axis_util-BczFISHz.js";
-import { m as f } from "./max-0Xnlpv8k.js";
-import { r as h } from "./reshape-BH7eBpwq.js";
-import { s as t } from "./sum-Cvq06317.js";
+import { q as l, w as c, E as y, a2 as E, p as w, a3 as o, a4 as u, l as v, f as I, a5 as A } from "./index-CamYe_M8.js";
+import { e as $ } from "./axis_util-CCNL7jea.js";
+import { m as f } from "./max-JBBv7aUf.js";
+import { r as h } from "./reshape-C45vIIRU.js";
+import { s as t } from "./sum-BpcpxNEh.js";
 /**
  * @license
  * Copyright 2020 Google Inc. All Rights Reserved.
@@ -40,21 +40,21 @@ const s = /* @__PURE__ */ l({ min_: k });
  * limitations under the License.
  * =============================================================================
  */
-function T(n, e = "euclidean", r = null, m = !1) {
+function q(n, e = "euclidean", r = null, m = !1) {
   n = c(n, "x", "norm");
-  const a = d(n, e, r);
+  const a = p(n, e, r);
   let i = a.shape;
   if (m) {
-    const p = w(r, n.shape);
-    i = $(a.shape, p);
+    const d = w(r, n.shape);
+    i = $(a.shape, d);
   }
   return h(a, i);
 }
-function d(n, e, r = null) {
+function p(n, e, r = null) {
   if (n.rank === 0)
     return o(n);
   if (n.rank !== 1 && r === null)
-    return d(h(n, [-1]), e, r);
+    return p(h(n, [-1]), e, r);
   if (n.rank === 1 || typeof r == "number" || Array.isArray(r) && r.length === 1) {
     if (e === 1)
       return t(o(n), r);
@@ -79,7 +79,7 @@ function d(n, e, r = null) {
   }
   throw new Error(`Error in norm: invalid axis: ${r}`);
 }
-const N = /* @__PURE__ */ l({ norm_: T });
+const N = /* @__PURE__ */ l({ norm_: q });
 export {
   s as m,
   N as n

package/dist/{ones-CrutWGas.js → ones-CMHNqMr6.js}

@@ -1,5 +1,5 @@
-import { n, p as t, q as m, E as i } from "./index-Du-bmOP8.js";
-import { z as c, c as f } from "./zeros-BaHhQTWf.js";
+import { y as n, B as t, i as m, E as i } from "./index-CamYe_M8.js";
+import { z as c, c as f } from "./zeros-DUkkVccu.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.

package/dist/ops/appendCache.js

@@ -1,8 +1,8 @@
-import { e as a } from "../index-Du-bmOP8.js";
+import { e as a } from "../index-CamYe_M8.js";
 import "./cpu/appendCache.js";
 import "./webgl/appendCache.js";
-import { c as s } from "../concat-DdKPyAtw.js";
-import { z as c } from "../zeros-BaHhQTWf.js";
+import { c as s } from "../concat-XOK9ANZu.js";
+import { z as c } from "../zeros-DUkkVccu.js";
 function i(r, p, n, o) {
   if (!o) {
     const e = r.shape[2];

package/dist/ops/attentionMask.js

@@ -1,4 +1,4 @@
-import { e as o } from "../index-Du-bmOP8.js";
+import { e as o } from "../index-CamYe_M8.js";
 import "./cpu/attentionMask.js";
 import "./webgl/attentionMask.js";
 import "./grads/attentionMask.js";

package/dist/ops/cpu/appendCache.js

@@ -1,5 +1,5 @@
-import { r as d } from "../../index-Du-bmOP8.js";
-import { c as h } from "../../concat-DdKPyAtw.js";
+import { r as d } from "../../index-CamYe_M8.js";
+import { c as h } from "../../concat-XOK9ANZu.js";
 function u(p) {
   const { cache: n, item: s } = p.inputs, { maxSize: r, pastLen: c } = p.attrs, t = n.shape[0], o = n.shape[1], a = n.shape[3], e = s.shape[2];
   if (c + e <= r) {

package/dist/ops/cpu/attentionMask.js

@@ -1,8 +1,8 @@
-import { r as a, g as p, f as u } from "../../index-Du-bmOP8.js";
-import { l as N, w as b } from "../../ops-CJNniCAV.js";
-import { o as g } from "../../ones-CrutWGas.js";
-import { z as A } from "../../zeros-BaHhQTWf.js";
-import { m as I } from "../../mat_mul-CbiqIe2d.js";
+import { r as a, g as p, f as u } from "../../index-CamYe_M8.js";
+import { l as N, w as b } from "../../ops-DqtYemmV.js";
+import { o as g } from "../../ones-CMHNqMr6.js";
+import { z as A } from "../../zeros-DUkkVccu.js";
+import { m as I } from "../../mat_mul-CAbRFWUj.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(g([t, t]), -1, 0).cast("bool"), l = A([t, t]), i = p([t, t], Number.NEGATIVE_INFINITY), f = b(m, l, i), k = I(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 { r as n } from "../../index-Du-bmOP8.js";
-import { s as f } from "../../softmax-DhWoBa7r.js";
+import { r as n } from "../../index-CamYe_M8.js";
+import { s as f } from "../../softmax-DvMvui-_.js";
 function r(t) {
   const { inputs: s, attrs: i } = t, { logits: o } = s, { dim: a, dropoutRate: e } = i;
   if (!o)

package/dist/ops/cpu/gatherSub.js

@@ -1,6 +1,6 @@
-import { o as u, j as c, E as g, O as h, r as m, s as p } from "../../index-Du-bmOP8.js";
-import { r as l } from "../../range-Bvs1hidm.js";
-import { s as N } from "../../stack-BV1v7l3S.js";
+import { q as u, w as c, E as g, Y as h, r as m, s as p } from "../../index-CamYe_M8.js";
+import { r as l } from "../../range-DqYjKnuG.js";
+import { s as N } from "../../stack-DgaoDmnF.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -23,8 +23,8 @@ function f(e, s) {
 }
 const b = /* @__PURE__ */ u({ gatherND_: f });
 function d(e) {
-  const { values: s, labels: n, logits: r } = e.inputs, t = n.shape[0], o = l(0, t, 1, "int32"), a = N([o, n], 1), i = b(r, a);
-  return p(s, i);
+  const { values: s, labels: n, logits: r } = e.inputs, t = n.shape[0], a = l(0, t, 1, "int32"), i = N([a, n], 1), o = b(r, i);
+  return p(s, o);
 }
 const k = {
   kernelName: "EfficientGatherSub",

package/dist/ops/cpu/gelu.js

@@ -1,4 +1,4 @@
-import { r as t, t as d } from "../../index-Du-bmOP8.js";
+import { r as t, t as d } from "../../index-CamYe_M8.js";
 const o = 0.7978845608028654, c = 0.044715;
 function m(u) {
   const { inputs: l } = u, { x: e } = l, n = e;

package/dist/ops/cpu/matMulGelu.js

@@ -1,4 +1,4 @@
-import { r as a, t as i } from "../../index-Du-bmOP8.js";
+import { r as a, t as i } from "../../index-CamYe_M8.js";
 const c = 0.7978845608028654, m = 0.044715;
 function M(o) {
   const { inputs: s } = o, { x: t, kernel: l } = s, e = t, u = l;

package/dist/ops/cpu/matMulMul.js

@@ -1,4 +1,4 @@
-import { r as n, t as M } from "../../index-Du-bmOP8.js";
+import { r as n, t as M } from "../../index-CamYe_M8.js";
 function e(t) {
   const { inputs: r, attrs: o } = t, { transposeA: s, transposeB: l } = o, { x: c, kernel: u, y: a } = r, m = c, i = u, k = a;
   return M(() => m.matMul(i, s, l).mul(k));

package/dist/ops/cpu/mulDropout.js

@@ -1,4 +1,4 @@
-import { r as e, b as u } from "../../index-Du-bmOP8.js";
+import { r as e, b as u } from "../../index-CamYe_M8.js";
 function n(o) {
   const { inputs: r } = o, { a: l, b: t } = r;
   return console.warn("Using fallback mulDrop implementation without dropout."), u(l, t);

package/dist/ops/cpu/normRMS.js

@@ -1,4 +1,4 @@
-import { r as o, t as d } from "../../index-Du-bmOP8.js";
+import { r as o, t as d } from "../../index-CamYe_M8.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 { r as q } from "../../index-Du-bmOP8.js";
-import { r as o } from "../../reshape-BH7eBpwq.js";
-import { s as x } from "../../split-BCUhuU7B.js";
+import { r as q } from "../../index-CamYe_M8.js";
+import { r as o } from "../../reshape-C45vIIRU.js";
+import { s as x } from "../../split-DxrHrPFK.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 { r as S } from "../../index-Du-bmOP8.js";
-import { r as F } from "../../range-Bvs1hidm.js";
-import { g as I } from "../../gather-D6MsdXqc.js";
-import { s as E } from "../../stack-BV1v7l3S.js";
-import { c as T } from "../../concat-DdKPyAtw.js";
+import { r as S } from "../../index-CamYe_M8.js";
+import { r as F } from "../../range-DqYjKnuG.js";
+import { g as I } from "../../gather-DKtUaTtA.js";
+import { s as E } from "../../stack-DgaoDmnF.js";
+import { c as T } from "../../concat-XOK9ANZu.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

@@ -1,39 +1,8 @@
-import { o as l, n as g, j as h, E as k, a5 as w, r as $, s as d, b as m } from "../../index-Du-bmOP8.js";
-import { r as b } from "../../range-Bvs1hidm.js";
-import { s as E } from "../../stack-BV1v7l3S.js";
-import { o as D } from "../../ones-CrutWGas.js";
-function N(a, r, t) {
-  const s = r.rank > 1 ? r.shape[r.rank - 1] : 1, e = r.rank > 1 ? r.rank - 1 : 1, o = `Must have updates.shape = indices.shape[:batchDim] + shape[sliceDim:], got updates.shape: ${t.shape}, indices.shape: ${r.shape}, shape: ${a}, sliceDim: ${s}, and batchDim: ${e}.`;
-  if (t.rank < e)
-    throw new Error(o + ` update.rank < ${e}. `);
-  if (a.length < s + (t.rank - e))
-    throw new Error(o + ` Output shape length < ${s + (t.rank - e)}`);
-  if (t.rank !== e + a.length - s)
-    throw new Error(o + ` update.rank != ${e + a.length - s}`);
-  for (let n = 0; n < e; ++n)
-    if (t.shape[n] !== r.shape[n])
-      throw new Error(o + ` updates.shape[${n}] (${t.shape[n]}) != indices.shape[${n}] (${r.shape[n]}).`);
-  for (let n = 0; n < t.rank - e; ++n)
-    if (t.shape[n + e] !== a[n + s])
-      throw new Error(o + ` updates.shape[${n + e}] (${t.shape[n + e]}) != shape[${n + e}] (${a[n + e]})`);
-}
-function S(a, r, t) {
-  if (r.rank < 1)
-    throw new Error(`tf.scatterND() expects the indices to be rank 1 or higher, but the rank was ${r.rank}.`);
-  if (a.rank < 1)
-    throw new Error(`tf.scatterND() expects the updates to be rank 1 or higher, but the rank was ${a.rank}.`);
-  if (r.dtype !== "int32")
-    throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${r.dtype}`);
-  if (t.length < 1)
-    throw new Error(`Output rank must be greater or equal to 1, but got shape: ${t}`);
-  if (t.length === 0) {
-    if (r.size === 0)
-      throw new Error(`Indices specified for empty output. indices shape: ${r.shape}`);
-    if (a.size === 0)
-      throw new Error(`Updates specified for empty output. updates shape: ${a.shape}`);
-  }
-  N(t, r, a);
-}
+import { q as f, y as g, w as o, E as l, X as N, r as b, s as S, b as h } from "../../index-CamYe_M8.js";
+import { v as D } from "../../scatter_nd_util-qgtnviTE.js";
+import { r as k } from "../../range-DqYjKnuG.js";
+import { s as v } from "../../stack-DgaoDmnF.js";
+import { o as E } from "../../ones-CMHNqMr6.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -50,21 +19,21 @@ function S(a, r, t) {
  * limitations under the License.
  * =============================================================================
  */
-function y(a, r, t) {
-  g(t);
-  const s = h(a, "indices", "scatterND", "int32"), e = h(r, "updates", "scatterND");
-  S(e, s, t);
-  const o = { indices: s, updates: e }, n = { shape: t };
-  return k.runKernel(w, o, n);
+function I(a, e, s) {
+  g(s);
+  const n = o(a, "indices", "scatterND", "int32"), t = o(e, "updates", "scatterND");
+  D(t, n, s);
+  const c = { indices: n, updates: t }, r = { shape: s };
+  return l.runKernel(N, c, r);
 }
-const v = /* @__PURE__ */ l({ scatterND_: y });
-function I(a) {
-  const { logits: r, labels: t, dy: s } = a.inputs, e = t.shape[0], o = r.shape[1], n = b(0, e, 1, "int32"), i = E([n, t], 1), c = D([e]), p = v(i, c, [e, o]), f = d(r, p), u = s.reshape([e, 1]);
-  return m(f, u);
+const y = /* @__PURE__ */ f({ scatterND_: I });
+function K(a) {
+  const { logits: e, labels: s, dy: n } = a.inputs, t = s.shape[0], c = e.shape[1], r = k(0, t, 1, "int32"), i = v([r, s], 1), d = E([t]), u = y(i, d, [t, c]), p = S(e, u), m = n.reshape([t, 1]);
+  return h(p, m);
 }
-const T = {
+const L = {
   kernelName: "EfficientScatterSub",
   backendName: "cpu",
-  kernelFunc: I
+  kernelFunc: K
 };
-$(T);
+b(L);

package/dist/ops/fusedSoftmax.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-Du-bmOP8.js";
+import { e as t } from "../index-CamYe_M8.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-Du-bmOP8.js";
+import { e as n } from "../index-CamYe_M8.js";
 import "./cpu/gatherSub.js";
 import "./webgl/gatherSub.js";
 function f(r, e, t) {

package/dist/ops/gelu.js

@@ -1,4 +1,4 @@
-import "../index-Du-bmOP8.js";
+import "../index-CamYe_M8.js";
 import "./cpu/gelu.js";
 import "./webgl/gelu.js";
 import { d as e, g as i } from "./grads/gelu.js";

package/dist/ops/grads/attentionMask.js

@@ -1,25 +1,29 @@
-import { h as l, f as a } from "../../index-Du-bmOP8.js";
-import { matMulMul as i } from "../matMulMul.js";
-const m = {
+import { h as m, f as i } from "../../index-CamYe_M8.js";
+import { matMulMul as u } from "../matMulMul.js";
+const p = {
   kernelName: "AttentionMask",
   inputsToSave: ["q", "k"],
   outputsToSave: [],
-  gradFunc: (t, u, c) => {
+  gradFunc: (t, c, l) => {
     if (Array.isArray(t))
       throw new Error("Expected dy to be a single Tensor");
-    const [e, o] = u, { divisor: n } = c;
+    const [e, n] = c, { divisor: a } = l;
     return {
-      q: () => i(t, o, a(n)),
+      q: () => u(t, n, i(a)),
       k: () => {
-        const s = e.transpose([0, 1, 3, 2]), r = i(s, t, a(n));
-        return s.dispose(), r.transpose([0, 1, 3, 2]);
+        const s = e.transpose([0, 1, 3, 2]), r = u(s, t, i(a));
+        s.dispose();
+        const o = r.transpose([0, 1, 3, 2]);
+        return r.dispose(), o;
       },
       mask: () => t,
       divisor: () => {
-        const s = e.matMul(o, !1, !0), r = t.mul(s);
-        return s.dispose(), r.sum();
+        const s = e.matMul(n, !1, !0), r = t.mul(s);
+        s.dispose();
+        const o = r.sum();
+        return r.dispose(), o;
       }
     };
   }
 };
-l(m);
+m(p);

package/dist/ops/grads/fusedSoftmax.js

@@ -1,20 +1,22 @@
-import { h as d, b as u, s as f } from "../../index-Du-bmOP8.js";
-import { mulDrop as l } from "../mulDrop.js";
-import { s as g } from "../../sum-Cvq06317.js";
-const T = {
+import { h as f, b as i, s as l } from "../../index-CamYe_M8.js";
+import { mulDrop as g } from "../mulDrop.js";
+import { s as T } from "../../sum-BpcpxNEh.js";
+const Y = {
   kernelName: "FusedSoftmax",
   outputsToSave: [!0],
-  gradFunc: (o, i, n) => {
-    const [s] = i, { dim: a, dropoutRate: t, seed: e } = n, p = !0, r = t && e ? l(o, s, t, e) : u(o, s);
+  gradFunc: (o, n, a) => {
+    const [s] = n, { dim: p, dropoutRate: t, seed: e } = a, c = !0, r = t && e ? g(o, s, t, e) : i(o, s);
     return {
       logits: () => {
-        const m = g(r, [a], p), c = u(m, s);
-        return m.dispose(), f(r, c);
+        const m = T(r, [p], c), u = i(m, s);
+        m.dispose();
+        const d = l(r, u);
+        return u.dispose(), d;
       }
     };
   }
 };
-d(T);
+f(Y);
 export {
-  T as softmaxGradConfig
+  Y as softmaxGradConfig
 };

package/dist/ops/grads/gelu.js

@@ -1,4 +1,4 @@
-import { h as t, e as n } from "../../index-Du-bmOP8.js";
+import { h as t, e as n } from "../../index-CamYe_M8.js";
 import "../cpu/gelu.js";
 import "../webgl/gelu.js";
 const o = {

package/dist/ops/grads/matMulGelu.js

@@ -1,4 +1,4 @@
-import { h as a, e as o } from "../../index-Du-bmOP8.js";
+import { h as a, e as o } from "../../index-CamYe_M8.js";
 function s(e, n, r) {
   return o().runKernel("MatMulGeluGrad", { dy: e, x: n, kernel: r });
 }

package/dist/ops/grads/normRMS.js

@@ -1,4 +1,4 @@
-import { h as t, e as g } from "../../index-Du-bmOP8.js";
+import { h as t, e as g } from "../../index-CamYe_M8.js";
 function s(r, a, n) {
   return g().runKernel("RMSNormGrad", { dy: r, x: a, gamma: n });
 }