@genai-fi/nanogpt 0.4.3 → 0.4.4

package/dist/layers/MLP.js

@@ -1,11 +1,11 @@
-import { t as _, c as M, e as d, H as v } from "../index-C4JCoBvj.js";
-import x from "./BaseLayer.js";
-import { g as L } from "../gelu-CnCt17Lk.js";
-import { v as n } from "../variable-LJT9Ld63.js";
-import { r as p, d as u } from "../dropout-DfDdklfL.js";
-import { r as l } from "../reshape-Boe4DuIO.js";
-import { m as f } from "../mat_mul-415y5Qn2.js";
-class V extends x {
+import { t as F, c as _, e as h, H as M } from "../index--6vO-cOz.js";
+import v from "./BaseLayer.js";
+import { matMulGelu as x } from "../ops/matMulGelu.js";
+import { v as c } from "../variable-BJTZ3jOy.js";
+import { r as d, d as u } from "../dropout-DFEXTPV0.js";
+import { r as p } from "../reshape-z51Eu-re.js";
+import { m as L } from "../mat_mul-BEHRPMh0.js";
+class G extends v {
   cFc = null;
   cProj = null;
   index;
@@ -15,12 +15,12 @@ class V extends x {
     super(s), this.index = t, this.hiddenUnits = s.gpt.mlpFactor * s.gpt.nEmbed;
   }
   build() {
-    this.cFc === null && (this.cFc = n(
-      p([this.config.gpt.nEmbed, this.hiddenUnits], 0, 0.02),
+    this.cFc === null && (this.cFc = c(
+      d([this.config.gpt.nEmbed, this.hiddenUnits], 0, 0.02),
       !0
       //`block_${this.index}_attn_cAttn_kernel`
-    )), this.cProj === null && (this.cProj = n(
-      p(
+    )), this.cProj === null && (this.cProj = c(
+      d(
         [this.hiddenUnits, this.config.gpt.nEmbed],
         0,
         0.02 / Math.sqrt(2 * this.config.gpt.nLayer)
@@ -45,43 +45,41 @@ class V extends x {
     const s = t.get(`block_${this.index}_mlpOut`)?.[0], i = t.get(`block_${this.index}_mlpHidden`)?.[0];
     if (!s || !i)
       throw new Error(`Weights for block ${this.index} not found`);
-    this.cFc ? this.cFc.assign(i) : this.cFc = n(i, !0), this.cProj ? this.cProj.assign(s) : this.cProj = n(s, !0);
+    this.cFc ? this.cFc.assign(i) : this.cFc = c(i, !0), this.cProj ? this.cProj.assign(s) : this.cProj = c(s, !0);
   }
   forward(t) {
-    return _(() => {
+    return F(() => {
       this.startMemory();
-      const [s, i, o] = t.shape, r = l(t, [s * i, o]), e = f(r, this.cFc), c = L(e);
+      const [s, i, r] = t.shape, o = p(t, [s * i, r]), e = x(o, this.cFc), n = L(e, this.cProj);
       e.dispose();
-      const a = f(c, this.cProj);
-      c.dispose();
-      const h = l(a, [s, i, o]);
-      return this.endMemory("MLP"), h;
+      const a = p(n, [s, i, r]);
+      return this.endMemory("MLP"), a;
     });
   }
   call(t, s = !1) {
     if (this.build(), s && this.config.layerConfig.checkpointMLP) {
-      const o = M(
+      const r = _(
         // @ts-expect-error Invalid params
-        (r, e, c, a) => {
-          const h = this.forward(r);
-          return a([r]), { value: h, gradFunc: (g, m) => {
-            const [b] = m, P = d().state.activeTape;
-            d().state.activeTape = [];
-            const j = v((F, w, T) => this.forward(F))([b, e, c], g);
-            return d().state.activeTape = P, j;
+        (o, e, n, a) => {
+          const l = this.forward(o);
+          return a([o]), { value: l, gradFunc: (f, g) => {
+            const [m] = g, b = h().state.activeTape;
+            h().state.activeTape = [];
+            const P = M((j, w, T) => this.forward(j))([m, e, n], f);
+            return h().state.activeTape = b, P;
           } };
         }
       )(t, this.cFc, this.cProj);
       if (this.config.gpt.dropout > 0) {
-        const r = u(o, this.config.gpt.dropout);
-        return o.dispose(), r;
+        const o = u(r, this.config.gpt.dropout);
+        return r.dispose(), o;
       }
-      return o;
+      return r;
     } else {
       const i = this.forward(t);
       if (s && this.config.gpt.dropout > 0) {
-        const o = u(i, this.config.gpt.dropout);
-        return i.dispose(), o;
+        const r = u(i, this.config.gpt.dropout);
+        return i.dispose(), r;
       }
       return i;
     }
@@ -91,5 +89,5 @@ class V extends x {
   }
 }
 export {
-  V as default
+  G as default
 };

package/dist/layers/RMSNorm.js

@@ -1,7 +1,7 @@
-import { t as r } from "../index-C4JCoBvj.js";
+import { t as r } from "../index--6vO-cOz.js";
 import m from "./BaseLayer.js";
-import { v as i } from "../variable-LJT9Ld63.js";
-import { o } from "../ones-Bf3YR48P.js";
+import { v as i } from "../variable-BJTZ3jOy.js";
+import { o } from "../ones-D6kB8bdY.js";
 class d extends m {
   gamma;
   epsilon;

package/dist/layers/RoPECache.js

@@ -1,6 +1,6 @@
-import { o as h, h as c, E as f, N as l, f as n, O as m, t as u, F as p } from "../index-C4JCoBvj.js";
-import { c as d, s as C } from "../sin-KmhiDuMa.js";
-import { r as a } from "../range-9AzeApCc.js";
+import { o as h, h as c, E as f, T as l, f as n, U as m, t as u, F as p } from "../index--6vO-cOz.js";
+import { c as d, s as C } from "../sin-H567uayl.js";
+import { r as a } from "../range-C_vpUjBu.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.

package/dist/layers/TiedEmbedding.js

@@ -1,8 +1,8 @@
-import { T as a } from "../TiedEmbedding-CnJ1bx4q.js";
-import "../index-C4JCoBvj.js";
-import "../tfjs_backend-Cug-PH75.js";
-import "../variable-LJT9Ld63.js";
-import "../gather-ZYRWhmXR.js";
+import { T as a } from "../TiedEmbedding-DznFwzcB.js";
+import "../index--6vO-cOz.js";
+import "../tfjs_backend-DuKis_xG.js";
+import "../variable-BJTZ3jOy.js";
+import "../gather-C5D8PxwA.js";
 export {
   a as default
 };

package/dist/layers/TransformerBlock.js

@@ -2,7 +2,7 @@ import h from "./CausalSelfAttention.js";
 import o from "./MLP.js";
 import a from "./RMSNorm.js";
 import p from "./BaseLayer.js";
-import { t as d } from "../index-C4JCoBvj.js";
+import { t as d } from "../index--6vO-cOz.js";
 class W extends p {
   ln1;
   attn;

package/dist/{log_sum_exp-BswFnwOb.js → log_sum_exp-CiEy1aUe.js}

@@ -1,8 +1,8 @@
-import { o as r, h as p, E as u, a3 as h, a4 as E, Y as S, s as $, a5 as d } from "./index-C4JCoBvj.js";
-import { e as K } from "./axis_util-BgTGy5w8.js";
-import { m as T } from "./max-CP_9O2Yd.js";
-import { r as m } from "./reshape-Boe4DuIO.js";
-import { s as _ } from "./sum-R28pucR5.js";
+import { o as r, h as p, E as u, a6 as h, a7 as E, $, s as S, a8 as d } from "./index--6vO-cOz.js";
+import { e as K } from "./axis_util-QP0LdI1v.js";
+import { m as T } from "./max-BUShNgfh.js";
+import { r as m } from "./reshape-z51Eu-re.js";
+import { s as _ } from "./sum-DdkDf2MG.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -62,12 +62,12 @@ const w = /* @__PURE__ */ r({ log_: v });
  * =============================================================================
  */
 function A(s, n = null, o = !1) {
-  const a = p(s, "x", "logSumExp"), t = S(n, a.shape), x = T(
+  const a = p(s, "x", "logSumExp"), t = $(n, a.shape), x = T(
     a,
     t,
     !0
     /* keepDims */
-  ), i = $(a, x), l = N(i), f = _(l, t), c = w(f), e = d(m(x, c.shape), c);
+  ), i = S(a, x), l = N(i), f = _(l, t), c = w(f), e = d(m(x, c.shape), c);
   if (o) {
     const g = K(e.shape, t);
     return m(e, g);

package/dist/main.js

@@ -1,11 +1,11 @@
-import { default as k } from "./NanoGPTModel.js";
-import { default as L } from "./TeachableLLM.js";
-import { default as b } from "./tokeniser/CharTokeniser.js";
-import { default as w } from "./tokeniser/bpe.js";
-import { default as D } from "./utilities/waitForModel.js";
-import { default as F } from "./data/textLoader.js";
-import { estimateMemoryUsage as N, estimateParameterCount as R, estimateResources as j, estimateTrainingMemoryUsage as q, validateConfig as z } from "./utilities/parameters.js";
-import "./index-C4JCoBvj.js";
+import { default as w } from "./NanoGPTModel.js";
+import { default as D } from "./TeachableLLM.js";
+import { default as F } from "./tokeniser/CharTokeniser.js";
+import { default as N } from "./tokeniser/bpe.js";
+import { default as j } from "./utilities/waitForModel.js";
+import { default as z } from "./data/textLoader.js";
+import { estimateMemoryUsage as H, estimateParameterCount as I, estimateResources as J, estimateTrainingMemoryUsage as K, validateConfig as O } from "./utilities/parameters.js";
+import "./index--6vO-cOz.js";
 import "./ops/cpu/scatterSub.js";
 import "./ops/webgl/scatterSub.js";
 import "./ops/cpu/gatherSub.js";
@@ -25,16 +25,22 @@ import "./ops/webgl/appendCache.js";
 import "./ops/cpu/fusedSoftmax.js";
 import "./ops/webgl/fusedSoftmax.js";
 import "./ops/grads/fusedSoftmax.js";
+import "./ops/cpu/matMulGelu.js";
+import "./ops/webgl/matMulGelu.js";
+import "./ops/grads/matMulGelu.js";
+import "./ops/cpu/gelu.js";
+import "./ops/webgl/gelu.js";
+import "./ops/grads/gelu.js";
 export {
-  w as BPETokeniser,
-  b as CharTokeniser,
-  k as NanoGPT,
-  L as TeachableLLM,
-  N as estimateMemoryUsage,
-  R as estimateParameterCount,
-  j as estimateResources,
-  q as estimateTrainingMemoryUsage,
-  F as loadTextData,
-  z as validateConfig,
-  D as waitForModel
+  N as BPETokeniser,
+  F as CharTokeniser,
+  w as NanoGPT,
+  D as TeachableLLM,
+  H as estimateMemoryUsage,
+  I as estimateParameterCount,
+  J as estimateResources,
+  K as estimateTrainingMemoryUsage,
+  z as loadTextData,
+  O as validateConfig,
+  j as waitForModel
 };

package/dist/{mat_mul-415y5Qn2.js → mat_mul-BEHRPMh0.js}

@@ -1,4 +1,4 @@
-import { o as m, h as s, p as c, E as M, B as p } from "./index-C4JCoBvj.js";
+import { o as m, h as s, p as c, E as M, B as p } from "./index--6vO-cOz.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.

package/dist/{max-CP_9O2Yd.js → max-BUShNgfh.js}

@@ -1,4 +1,4 @@
-import { o as r, h as e, E as x, M as c } from "./index-C4JCoBvj.js";
+import { o as r, h as e, E as x, M as c } from "./index--6vO-cOz.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.

package/dist/{moments-CjeIaVdp.js → moments-DYOHXoRV.js}

@@ -1,6 +1,6 @@
-import { o as m, h as c, E as f, X as i, Y as l, Z as h, s as x, x as d } from "./index-C4JCoBvj.js";
-import { e as v } from "./axis_util-BgTGy5w8.js";
-import { r as E } from "./reshape-Boe4DuIO.js";
+import { o as m, h as c, E as f, _ as i, $ as l, a0 as h, s as x, x as d } from "./index--6vO-cOz.js";
+import { e as v } from "./axis_util-QP0LdI1v.js";
+import { r as E } from "./reshape-z51Eu-re.js";
 /**
  * @license
  * Copyright 2020 Google Inc. All Rights Reserved.
@@ -46,8 +46,8 @@ function T(a, t = null, e = !1) {
   const p = h(x(d(a, "float32"), E(s, o))), u = r(p, n, e);
   return { mean: s, variance: u };
 }
-const N = /* @__PURE__ */ m({ moments_: T });
+const K = /* @__PURE__ */ m({ moments_: T });
 export {
-  N as a,
+  K as a,
   r as m
 };

package/dist/{norm-CZM380I3.js → norm-DSva3hI3.js}

@@ -1,8 +1,8 @@
-import { o as l, h as c, E as y, _ as E, Y as w, $ as o, a0 as u, O as v, f as I, Z as $ } from "./index-C4JCoBvj.js";
-import { e as A } from "./axis_util-BgTGy5w8.js";
-import { m as f } from "./max-CP_9O2Yd.js";
-import { r as h } from "./reshape-Boe4DuIO.js";
-import { s as t } from "./sum-R28pucR5.js";
+import { o as l, h as c, E as y, a1 as E, $ as w, a2 as o, a3 as u, U as v, f as I, a0 as $ } from "./index--6vO-cOz.js";
+import { e as A } from "./axis_util-QP0LdI1v.js";
+import { m as f } from "./max-BUShNgfh.js";
+import { r as h } from "./reshape-z51Eu-re.js";
+import { s as t } from "./sum-DdkDf2MG.js";
 /**
  * @license
  * Copyright 2020 Google Inc. All Rights Reserved.
@@ -20,8 +20,8 @@ import { s as t } from "./sum-R28pucR5.js";
  * =============================================================================
  */
 function k(n, e = null, r = !1) {
-  const i = { x: c(n, "x", "min") }, a = { axis: e, keepDims: r };
-  return y.runKernel(E, i, a);
+  const a = { x: c(n, "x", "min") }, i = { axis: e, keepDims: r };
+  return y.runKernel(E, a, i);
 }
 const s = /* @__PURE__ */ l({ min_: k });
 /**
@@ -42,13 +42,13 @@ const s = /* @__PURE__ */ l({ min_: k });
  */
 function T(n, e = "euclidean", r = null, m = !1) {
   n = c(n, "x", "norm");
-  const i = d(n, e, r);
-  let a = i.shape;
+  const a = d(n, e, r);
+  let i = a.shape;
   if (m) {
     const p = w(r, n.shape);
-    a = A(i.shape, p);
+    i = A(a.shape, p);
   }
-  return h(i, a);
+  return h(a, i);
 }
 function d(n, e, r = null) {
   if (n.rank === 0)
@@ -79,8 +79,8 @@ function d(n, e, r = null) {
   }
   throw new Error(`Error in norm: invalid axis: ${r}`);
 }
-const K = /* @__PURE__ */ l({ norm_: T });
+const N = /* @__PURE__ */ l({ norm_: T });
 export {
   s as m,
-  K as n
+  N as n
 };

package/dist/{ones-Bf3YR48P.js → ones-D6kB8bdY.js}

@@ -1,5 +1,5 @@
-import { k as n, l as t, n as m, E as i } from "./index-C4JCoBvj.js";
-import { z as l, c } from "./zeros-dnQxFgAD.js";
+import { k as n, l as t, n as m, E as i } from "./index--6vO-cOz.js";
+import { z as l, c } from "./zeros-8xl-W2DC.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.

package/dist/ops/appendCache.js

@@ -1,8 +1,8 @@
-import { e as a } from "../index-C4JCoBvj.js";
+import { e as a } from "../index--6vO-cOz.js";
 import "./cpu/appendCache.js";
 import "./webgl/appendCache.js";
-import { z as s } from "../zeros-dnQxFgAD.js";
-import { c } from "../concat-CuRsVY-K.js";
+import { z as s } from "../zeros-8xl-W2DC.js";
+import { c } from "../concat-DvWM7HGZ.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 i } from "../index-C4JCoBvj.js";
+import { e as i } from "../index--6vO-cOz.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-C4JCoBvj.js";
-import { c as h } from "../../concat-CuRsVY-K.js";
+import { r as d } from "../../index--6vO-cOz.js";
+import { c as h } from "../../concat-DvWM7HGZ.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

@@ -1,5 +1,5 @@
-import { r as o, f as k } from "../../index-C4JCoBvj.js";
-import { m as d } from "../../mat_mul-415y5Qn2.js";
+import { r as o, f as k } from "../../index--6vO-cOz.js";
+import { m as d } from "../../mat_mul-BEHRPMh0.js";
 function r(t) {
   const { q: e, k: n, mask: s } = t.inputs, { divisor: c } = t.attrs, m = e.shape[2], i = n.shape[2], a = d(e, n, !1, !0).mul(k(c));
   if (s) {

package/dist/ops/cpu/fusedSoftmax.js

@@ -1,5 +1,5 @@
-import { r as n } from "../../index-C4JCoBvj.js";
-import { s as f } from "../../softmax-Cujsg4ay.js";
+import { r as n } from "../../index--6vO-cOz.js";
+import { s as f } from "../../softmax-Dsxflvdl.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, h as c, E as g, L as h, r as m, s as p } from "../../index-C4JCoBvj.js";
-import { r as l } from "../../range-9AzeApCc.js";
-import { s as N } from "../../stack-D1YjmgKN.js";
+import { o as u, h as c, E as g, N as h, r as m, s as p } from "../../index--6vO-cOz.js";
+import { r as N } from "../../range-C_vpUjBu.js";
+import { s as l } from "../../stack-CmqSdsfs.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -23,7 +23,7 @@ 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);
+  const { values: s, labels: n, logits: r } = e.inputs, t = n.shape[0], o = N(0, t, 1, "int32"), a = l([o, n], 1), i = b(r, a);
   return p(s, i);
 }
 const k = {

package/dist/ops/cpu/gelu.js

@@ -1,4 +1,4 @@
-import { r as t, t as d } from "../../index-C4JCoBvj.js";
+import { r as t, t as d } from "../../index--6vO-cOz.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.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/ops/cpu/matMulGelu.js

@@ -0,0 +1,40 @@
+import { r as a, t as i } from "../../index--6vO-cOz.js";
+const c = 0.7978845608028654, m = 0.044715;
+function M(o) {
+  const { inputs: s } = o, { x: t, kernel: l } = s, e = t, u = l;
+  return i(() => {
+    const n = e.matMul(u), d = n.mul(n).mul(n), r = n.add(d.mul(m)).mul(c).tanh().add(1).mul(0.5);
+    return n.mul(r);
+  });
+}
+const b = {
+  kernelName: "MatMulGelu",
+  backendName: "cpu",
+  kernelFunc: M
+};
+a(b);
+const w = {
+  kernelName: "MatMulGelu",
+  backendName: "tensorflow",
+  kernelFunc: M
+};
+a(w);
+function f(o) {
+  const { dy: s, x: t, kernel: l } = o.inputs;
+  return i(() => {
+    const e = t.matMul(l), u = e.square(), n = u.mul(e), r = e.add(n.mul(m)).mul(c).tanh(), G = r.square().neg().add(1), g = u.mul(3 * m).add(1), p = r.add(1).mul(0.5), N = e.mul(G).mul(c).mul(g).mul(0.5), K = p.add(N), k = s.mul(K), h = k.matMul(l.transpose()), F = t.transpose().matMul(k);
+    return [h, F];
+  });
+}
+const x = {
+  kernelName: "MatMulGeluGrad",
+  backendName: "cpu",
+  kernelFunc: f
+};
+a(x);
+const C = {
+  kernelName: "MatMulGeluGrad",
+  backendName: "tensorflow",
+  kernelFunc: f
+};
+a(C);

package/dist/ops/cpu/mulDropout.js

@@ -1,4 +1,4 @@
-import { r as e, b as u } from "../../index-C4JCoBvj.js";
+import { r as e, b as u } from "../../index--6vO-cOz.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/qkv.js

@@ -1,6 +1,6 @@
-import { r as q } from "../../index-C4JCoBvj.js";
-import { r as o } from "../../reshape-Boe4DuIO.js";
-import { s as x } from "../../split-DbcNm1-i.js";
+import { r as q } from "../../index--6vO-cOz.js";
+import { r as o } from "../../reshape-z51Eu-re.js";
+import { s as x } from "../../split-B_k_jwud.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-C4JCoBvj.js";
-import { r as F } from "../../range-9AzeApCc.js";
-import { g as I } from "../../gather-ZYRWhmXR.js";
-import { s as E } from "../../stack-D1YjmgKN.js";
-import { c as T } from "../../concat-CuRsVY-K.js";
+import { r as S } from "../../index--6vO-cOz.js";
+import { r as F } from "../../range-C_vpUjBu.js";
+import { g as I } from "../../gather-C5D8PxwA.js";
+import { s as E } from "../../stack-CmqSdsfs.js";
+import { c as T } from "../../concat-DvWM7HGZ.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

@@ -1,7 +1,7 @@
-import { o as l, k, h, E as g, a2 as w, r as $, s as d, b as m } from "../../index-C4JCoBvj.js";
-import { r as b } from "../../range-9AzeApCc.js";
-import { s as E } from "../../stack-D1YjmgKN.js";
-import { o as D } from "../../ones-Bf3YR48P.js";
+import { o as l, k, h, E as g, a5 as w, r as $, s as d, b as m } from "../../index--6vO-cOz.js";
+import { r as b } from "../../range-C_vpUjBu.js";
+import { s as E } from "../../stack-CmqSdsfs.js";
+import { o as D } from "../../ones-D6kB8bdY.js";
 function N(n, 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: ${n}, sliceDim: ${s}, and batchDim: ${e}.`;
   if (t.rank < e)

package/dist/ops/fusedSoftmax.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-C4JCoBvj.js";
+import { e as t } from "../index--6vO-cOz.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-C4JCoBvj.js";
+import { e as n } from "../index--6vO-cOz.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-C4JCoBvj.js";
+import "../index--6vO-cOz.js";
 import "./cpu/gelu.js";
 import "./webgl/gelu.js";
-import { d as e, g as i } from "../gelu-CnCt17Lk.js";
+import { d as e, g as i } from "./grads/gelu.js";
 export {
   e as dGelu,
   i as gelu

package/dist/ops/grads/attentionMask.js

@@ -1,4 +1,4 @@
-import { g as i } from "../../index-C4JCoBvj.js";
+import { g as i } from "../../index--6vO-cOz.js";
 const u = {
   kernelName: "AttentionMask",
   inputsToSave: ["q", "k"],

package/dist/ops/grads/fusedSoftmax.js

@@ -1,6 +1,6 @@
-import { g as p, b as m, s as d } from "../../index-C4JCoBvj.js";
+import { g as p, b as m, s as d } from "../../index--6vO-cOz.js";
 import { mulDrop as c } from "../mulDrop.js";
-import { s as f } from "../../sum-R28pucR5.js";
+import { s as f } from "../../sum-DdkDf2MG.js";
 const g = {
   kernelName: "FusedSoftmax",
   outputsToSave: [!0],

package/dist/ops/grads/gelu.js

@@ -1,5 +1,26 @@
-import "../../index-C4JCoBvj.js";
-import { a as e } from "../../gelu-CnCt17Lk.js";
+import { g as t, e as n } from "../../index--6vO-cOz.js";
+import "../cpu/gelu.js";
+import "../webgl/gelu.js";
+const o = {
+  kernelName: "Gelu",
+  inputsToSave: ["x"],
+  outputsToSave: [],
+  gradFunc: (e, r) => {
+    const [u] = r;
+    return {
+      x: () => a(e, u)
+    };
+  }
+};
+t(o);
+function g(e) {
+  return n().runKernel("Gelu", { x: e });
+}
+function a(e, r) {
+  return n().runKernel("GeluGrad", { dy: e, x: r });
+}
 export {
-  e as geluGradConfig
+  a as d,
+  g,
+  o as geluGradConfig
 };

package/dist/ops/grads/matMulGelu.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/ops/grads/matMulGelu.js

@@ -0,0 +1,17 @@
+import { g as a, e as o } from "../../index--6vO-cOz.js";
+function s(e, n, r) {
+  return o().runKernel("MatMulGeluGrad", { dy: e, x: n, kernel: r });
+}
+const d = {
+  kernelName: "MatMulGelu",
+  inputsToSave: ["x", "kernel"],
+  outputsToSave: [],
+  gradFunc: (e, n) => {
+    const [r, t] = n, [u, l] = s(e, r, t);
+    return {
+      x: () => u,
+      kernel: () => l
+    };
+  }
+};
+a(d);

package/dist/ops/grads/qkv.js

@@ -1,4 +1,4 @@
-import { g as v } from "../../index-C4JCoBvj.js";
+import { g as v } from "../../index--6vO-cOz.js";
 const g = {
   kernelName: "QKV",
   inputsToSave: ["x", "kernel"],

package/dist/ops/grads/rope.js

@@ -1,4 +1,4 @@
-import { g as a, e as i } from "../../index-C4JCoBvj.js";
+import { g as a, e as i } from "../../index--6vO-cOz.js";
 function p(n, e, s, o) {
   return i().runKernel("Rope", { x: n, sin: e, cos: s }, { pastLen: o });
 }

package/dist/ops/matMulGelu.d.ts

@@ -0,0 +1,3 @@
+import { Tensor } from '@tensorflow/tfjs-core';
+export declare function matMulGelu(x: Tensor, kernel: Tensor): Tensor;
+export declare function dMatMulGelu(dy: Tensor, x: Tensor, kernel: Tensor): Tensor;

package/dist/ops/matMulGelu.js

@@ -0,0 +1,14 @@
+import { e as u } from "../index--6vO-cOz.js";
+import "./cpu/matMulGelu.js";
+import "./webgl/matMulGelu.js";
+import "./grads/matMulGelu.js";
+function M(r, e) {
+  return u().runKernel("MatMulGelu", { x: r, kernel: e });
+}
+function a(r, e, t) {
+  return u().runKernel("MatMulGeluGrad", { dy: r, x: e, kernel: t });
+}
+export {
+  a as dMatMulGelu,
+  M as matMulGelu
+};