@genai-fi/nanogpt 0.19.0 → 0.20.0

package/dist/ops/dot16.js

@@ -1,42 +0,0 @@
-import { b as d } from "../matMul16-BcVC_E62.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-BGnG-ppu.js";
-function E(e, s, h = !1, c = !1) {
-  if (!p(e) && !p(s))
-    return x(e, s);
-  if (e.rank < 2 || s.rank < 2)
-    throw new Error(
-      `dot requires both inputs to be rank >= 2 but got x shape = ${e.shape} and y shape = ${s.shape}`
-    );
-  if (s.rank >= 3) {
-    const r = e.shape.slice(-1)[0], i = s.shape.slice(-2)[0];
-    if (r !== i)
-      throw new Error(
-        `If rank y >= 3, then the second last dim of y must equal the last dim of x but got x shape = ${e.shape} and  y shape = ${s.shape}`
-      );
-  }
-  if (!p(e) || !p(s))
-    throw new Error("dot16 requires both inputs to be packed Tensors.");
-  if (e.rank === 2 && s.rank === 2)
-    return d(e, s, h, c);
-  {
-    const r = e.shape.slice(), i = r.pop();
-    e = n(e, [-1, i]);
-    const a = s.shape.slice(), l = a.pop(), m = a.pop(), k = [...a, l], f = Array.from({ length: s.rank }, (o, t) => t === 0 ? s.rank - 2 : t <= s.rank - 2 ? t - 1 : t);
-    if (f.every((o, t) => o === t))
-      s = n(s, [m, -1]);
-    else {
-      const o = w(s, f);
-      s = n(o, [m, -1]), o.dispose();
-    }
-    const y = [...r, ...k], u = d(e, s, h, c);
-    e.dispose(), s.dispose();
-    const D = n(u, y);
-    return u.dispose(), D;
-  }
-}
-export {
-  E as dot16
-};

package/dist/ops/dropout.d.ts

@@ -1,2 +0,0 @@
-import { Tensor, TensorLike } from '@tensorflow/tfjs-core';
-export declare function dropout(x: Tensor | TensorLike, rate: number, noiseShape?: number[], seed?: number | string): Tensor;

package/dist/ops/dropout.js

@@ -1,14 +0,0 @@
-import { q as m, x as f, ag as p, i as u, j as c } from "../index-CUXkjxiT.js";
-import { g as l, r as d } from "../dropout_util--NxWuYg2.js";
-import { mul16 as g } from "./mul16.js";
-import { f as b } from "../floor-BQtb-Azg.js";
-function v(t, o, n, e) {
-  const r = m(t, "x", "dropout");
-  if (f(o >= 0 && o < 1, () => `rate must be a float in the range [0, 1), but got ${o}.`), o === 0)
-    return t instanceof p ? r.clone() : r;
-  const a = l(r, n), s = 1 - o, i = u(b(c(d(a, 0, 1, "float32", e), s)), s);
-  return g(r, i);
-}
-export {
-  v as dropout
-};

package/dist/ops/dropout16.d.ts

@@ -1,2 +0,0 @@
-import { Rank, Tensor } from '@tensorflow/tfjs-core';
-export declare function dropout16<R extends Rank = Rank>(x: Tensor<R>, dropout: number, seed?: number): Tensor<R>;

package/dist/ops/dropout16.js

@@ -1,25 +0,0 @@
-import { isPackedTensor as u } from "../utilities/packed.js";
-import { e as d, h as i } from "../index-CUXkjxiT.js";
-import "./webgl/dropout16.js";
-const s = {
-  kernelName: "Dropout16",
-  inputsToSave: [],
-  gradFunc: (r, o, e) => {
-    const { dropout: t, seed: n } = e;
-    return { x: () => a(r, t, n) };
-  }
-};
-d(s);
-function a(r, o, e) {
-  return u(r) ? i().runKernel(
-    "Dropout16",
-    { x: r },
-    {
-      dropout: o,
-      seed: e ?? Math.random()
-    }
-  ) : r;
-}
-export {
-  a as dropout16
-};

package/dist/ops/gatherSub.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function gatherSub(values: Tensor, labels: Tensor, logits: Tensor): Tensor;

package/dist/ops/gatherSub.js

@@ -1,9 +0,0 @@
-import { h as n } from "../index-CUXkjxiT.js";
-import "./cpu/gatherSub.js";
-import "./webgl/gatherSub.js";
-function f(r, e, t) {
-  return n().runKernel("EfficientGatherSub", { logits: t, labels: e, values: r }, {});
-}
-export {
-  f as gatherSub
-};

package/dist/ops/gelu.d.ts

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

package/dist/ops/gelu.js

@@ -1,8 +0,0 @@
-import "../index-CUXkjxiT.js";
-import "./cpu/gelu.js";
-import "./webgl/gelu.js";
-import { d as e, g as i } from "../gelu-B220X1Go.js";
-export {
-  e as dGelu,
-  i as gelu
-};

package/dist/ops/globalNorm.d.ts

@@ -1,2 +0,0 @@
-import { Scalar, Tensor } from '@tensorflow/tfjs-core';
-export declare function clipScale(grads: Tensor[], invLossScaling: number, clipNorm?: number): Scalar;

package/dist/ops/globalNorm.js

@@ -1,13 +0,0 @@
-import { t as i, h as n } from "../index-CUXkjxiT.js";
-import { z as l } from "../zeros-DvZpK8s6.js";
-function c(r, e, o = 1) {
-  return i(() => {
-    const t = l([r.length], "int32");
-    return r.map(
-      (p, u) => n().runKernel("Norm2", { x: p, output: t }, { invLossScaling: e, index: u })
-    ), n().runKernel("ClipScale", { x: t }, { invLossScaling: e, clipNorm: o });
-  });
-}
-export {
-  c as clipScale
-};

package/dist/ops/grads/add16.d.ts

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

package/dist/ops/grads/add16.js

@@ -1,26 +0,0 @@
-import { e as u, _ as h, $ as d } from "../../index-CUXkjxiT.js";
-import { sum16 as c } from "../sum16.js";
-import { reshape16 as p } from "../reshape16.js";
-const A = {
-  kernelName: "Add16",
-  inputsToSave: ["a", "b"],
-  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: () => {
-      let e = s;
-      const r = d(t.shape, n);
-      r.length > 0 && (e = c(e, r));
-      const o = p(e, t.shape);
-      return e.dispose(), o;
-    }, b: () => {
-      let e = s;
-      const r = d(a.shape, n);
-      r.length > 0 && (e = c(e, r));
-      const o = p(e, a.shape);
-      return e.dispose(), o;
-    } };
-  }
-};
-u(A);

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

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

package/dist/ops/grads/attentionMask.js

@@ -1,21 +0,0 @@
-import { e as m } from "../../index-CUXkjxiT.js";
-import { m as o } from "../../matMul16-BcVC_E62.js";
-import { transpose16 as c } from "../transpose16.js";
-const l = {
-  kernelName: "AttentionMask",
-  inputsToSave: ["q", "k"],
-  outputsToSave: [],
-  gradFunc: (r, s, n) => {
-    if (Array.isArray(r))
-      throw new Error("Expected dy to be a single Tensor");
-    const [a, i] = s, { divisor: e } = n;
-    return {
-      q: () => o(r, i, e),
-      k: () => {
-        const t = o(a, r, e, !0, !1), u = c(t, [0, 1, 3, 2]);
-        return t.dispose(), u;
-      }
-    };
-  }
-};
-m(l);

package/dist/ops/grads/dropout16.d.ts

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

package/dist/ops/grads/dropout16.js

@@ -1,2 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import "../dropout16.js";

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

@@ -1,2 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const geluGradConfig: GradConfig;

package/dist/ops/grads/gelu.js

@@ -1,5 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import { a as e } from "../../gelu-B220X1Go.js";
-export {
-  e as geluGradConfig
-};

package/dist/ops/grads/matMul16.d.ts

@@ -1,2 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const matMul16GradConfig: GradConfig;

package/dist/ops/grads/matMul16.js

@@ -1,9 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import { a as f } from "../../matMul16-BcVC_E62.js";
-import "../../gelu-B220X1Go.js";
-import "../transpose16.js";
-import "../reshape16.js";
-import "../mul16.js";
-export {
-  f as matMul16GradConfig
-};

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

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

package/dist/ops/grads/matMulGelu.js

@@ -1,17 +0,0 @@
-import { e as a, h as o } from "../../index-CUXkjxiT.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/mul16.d.ts

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

package/dist/ops/grads/mul16.js

@@ -1,4 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import "../mul16.js";
-import "../sum16.js";
-import "../reshape16.js";

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

@@ -1,3 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const normRMSGradConfig: GradConfig;
-export declare const normRMSGradConfigNoGamma: GradConfig;

package/dist/ops/grads/normRMS.js

@@ -1,33 +0,0 @@
-import { e, h as u } from "../../index-CUXkjxiT.js";
-function m(r, n, o) {
-  return u().runKernel("RMSNormGrad", o ? { dy: r, x: n, gamma: o } : { dy: r, x: n });
-}
-const S = {
-  kernelName: "RMSNorm",
-  inputsToSave: ["x", "gamma"],
-  outputsToSave: [],
-  gradFunc: (r, n) => {
-    const [o, a] = n, [t, s] = m(r, o, a);
-    return {
-      x: () => t,
-      gamma: () => s
-    };
-  }
-};
-e(S);
-const c = {
-  kernelName: "RMSNormNoGamma",
-  inputsToSave: ["x"],
-  outputsToSave: [],
-  gradFunc: (r, n) => {
-    const [o] = n, [a] = m(r, o);
-    return {
-      x: () => a
-    };
-  }
-};
-e(c);
-export {
-  S as normRMSGradConfig,
-  c as normRMSGradConfigNoGamma
-};

package/dist/ops/grads/pack16.d.ts

@@ -1,2 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const packGradConfig: GradConfig;

package/dist/ops/grads/pack16.js

@@ -1,6 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import { b as i } from "../../pack16-bqltoUlR.js";
-import "../../slice-BzS11Qh0.js";
-export {
-  i as packGradConfig
-};

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

@@ -1,3 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-import { NamedGradientMap } from '@tensorflow/tfjs-core/dist/tape';
-export declare function qkvGrad(dy: Tensor[], x: Tensor, kernel: Tensor): NamedGradientMap;

package/dist/ops/grads/qkv.js

@@ -1,34 +0,0 @@
-import { e as u } from "../../index-CUXkjxiT.js";
-import { a as f } from "../../matMul16-BcVC_E62.js";
-import { concat16 as g } from "../concat16.js";
-import { sum16 as l } from "../sum16.js";
-import { s as k } from "../../squeeze-ho4wLUek.js";
-const i = {
-  kernelName: "QKV",
-  inputsToSave: ["x", "kernel"],
-  outputsToSave: [],
-  gradFunc: (o, s) => {
-    const [r, n, t] = o, [a] = s, p = g([r, n, t], 1);
-    r.dispose(), n.dispose(), t.dispose();
-    const m = [a.shape[0], a.shape[1], 3 * a.shape[2]], d = f.gradFunc(p, s, {
-      transposeA: !1,
-      transposeB: !1,
-      originalShape: m,
-      perm: [0, 2, 1, 3]
-    });
-    return p.dispose(), {
-      x: () => d.A(),
-      kernel: () => {
-        const e = d.B(), c = e.shape[0] === 1 ? k(e, [0]) : l(e, 0);
-        return e.dispose(), c;
-      }
-    };
-  }
-};
-function B(o, s, r) {
-  return i.gradFunc(o, [s, r], {});
-}
-u(i);
-export {
-  B as qkvGrad
-};

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

@@ -1,2 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const ropeGradConfig: GradConfig;

package/dist/ops/grads/rope.js

@@ -1,5 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import { a as p } from "../../rope-DcrZM_e6.js";
-export {
-  p as ropeGradConfig
-};

package/dist/ops/grads/softmax16.d.ts

@@ -1,2 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const softmax16GradConfig: GradConfig;

package/dist/ops/grads/softmax16.js

@@ -1,25 +0,0 @@
-import { e as n, h as a } from "../../index-CUXkjxiT.js";
-import { isPackedTensor as t } from "../../utilities/packed.js";
-function s(r, e) {
-  return a().runKernel("Softmax16Grad", { dy: r, softmaxOutput: e });
-}
-const i = {
-  kernelName: "Softmax16",
-  outputsToSave: [!0],
-  gradFunc: (r, e) => {
-    const [o] = e;
-    if (Array.isArray(r))
-      throw new Error("Expected dy to be a single Tensor");
-    if (!t(o))
-      throw console.error(o), new Error("Softmax16 gradient requires packed y Tensor");
-    if (!t(r))
-      throw new Error("Softmax16 gradient requires packed dy Tensor");
-    return {
-      logits: () => s(r, o)
-    };
-  }
-};
-n(i);
-export {
-  i as softmax16GradConfig
-};

package/dist/ops/grads/unpack16.d.ts

@@ -1,2 +0,0 @@
-import { GradConfig } from '@tensorflow/tfjs-core';
-export declare const unpackGradConfig: GradConfig;

package/dist/ops/grads/unpack16.js

@@ -1,5 +0,0 @@
-import "../../index-CUXkjxiT.js";
-import { a as p } from "../../pack16-bqltoUlR.js";
-export {
-  p as unpackGradConfig
-};

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

@@ -1,4 +0,0 @@
-import { TensorInfo } from '@tensorflow/tfjs-core';
-export declare function forceFloat<T extends TensorInfo>(x: T): T;
-export declare function forceInt<T extends TensorInfo>(x: T): T;
-export declare function forcePacked<T extends TensorInfo>(x: T): T;

package/dist/ops/grads/utils.js

@@ -1,14 +0,0 @@
-function e(t) {
-  return t.dtype = "float32", t;
-}
-function n(t) {
-  return t.dtype = "int32", t;
-}
-function r(t) {
-  return t.dtype = "packedF16", t;
-}
-export {
-  e as forceFloat,
-  n as forceInt,
-  r as forcePacked
-};

package/dist/ops/log.js

@@ -1 +0,0 @@
-import "./webgl/log.js";

package/dist/ops/matMul16.d.ts

@@ -1,15 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function matMul16(A: Tensor, B: Tensor, transposeA?: boolean, transposeB?: boolean, attrs?: {
-    scale?: number;
-    scaleA?: number;
-    scaleB?: number;
-    activation?: 'gelu' | 'relu2' | 'relu';
-    forceOutputShape?: number[];
-    perm?: number[];
-    causalMask?: boolean;
-    pastLen?: number;
-}): Tensor;
-export declare function matMul16Scaled(A: Tensor, B: Tensor, scale: number, transposeA?: boolean, transposeB?: boolean): Tensor;
-export declare function matMul16ScaleA(A: Tensor, B: Tensor, scale: number, transposeA?: boolean, transposeB?: boolean): Tensor;
-export declare function matMul16ScaleB(A: Tensor, B: Tensor, scale: number, transposeA?: boolean, transposeB?: boolean): Tensor;
-export declare function matMul16Gelu(A: Tensor, B: Tensor, transposeA?: boolean, transposeB?: boolean): Tensor;

package/dist/ops/matMul16.js

@@ -1,13 +0,0 @@
-import "../index-CUXkjxiT.js";
-import { b as p, c as u, d as i, e as s, m as M } from "../matMul16-BcVC_E62.js";
-import "./webgl/matMul16.js";
-import "./cpu/matMul16.js";
-import "../utilities/packed.js";
-import "../pack16-bqltoUlR.js";
-export {
-  p as matMul16,
-  u as matMul16Gelu,
-  i as matMul16ScaleA,
-  s as matMul16ScaleB,
-  M as matMul16Scaled
-};

package/dist/ops/matMulGelu.d.ts

@@ -1,3 +0,0 @@
-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

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

package/dist/ops/matMulMul.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function matMulMul(x: Tensor, kernel: Tensor, y: Tensor, transposeA?: boolean, transposeB?: boolean): Tensor;

package/dist/ops/matMulMul.js

@@ -1,9 +0,0 @@
-import { h as u } from "../index-CUXkjxiT.js";
-import "./cpu/matMulMul.js";
-import "./webgl/matMulMul.js";
-function m(r, e, t, l = !1, n = !1) {
-  return u().runKernel("MatMulMul", { x: r, kernel: e, y: t }, { transposeA: l, transposeB: n });
-}
-export {
-  m as matMulMul
-};

package/dist/ops/mul16.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function mul16(a: Tensor, b: Tensor): Tensor;

package/dist/ops/mul16.js

@@ -1,39 +0,0 @@
-import { e as h, _ as f, $ as d, m as g, h as A } from "../index-CUXkjxiT.js";
-import { isPackedTensor as i } from "../utilities/packed.js";
-import { sum16 as p } from "./sum16.js";
-import { reshape16 as l } from "./reshape16.js";
-const x = {
-  kernelName: "Mul16",
-  inputsToSave: ["a", "b"],
-  gradFunc: (r, s) => {
-    const [t, o] = s, c = f(t.shape, o.shape);
-    if (Array.isArray(r))
-      throw new Error("Mul16 gradFunc expected dy to be a Tensor but got an array");
-    return { a: () => {
-      let e = m(r, o);
-      const n = d(t.shape, c);
-      if (n.length > 0) {
-        const u = p(e, n);
-        e.dispose(), e = u;
-      }
-      const a = l(e, t.shape);
-      return e.dispose(), a;
-    }, b: () => {
-      let e = m(r, t);
-      const n = d(o.shape, c);
-      if (n.length > 0) {
-        const u = p(e, n);
-        e.dispose(), e = u;
-      }
-      const a = l(e, o.shape);
-      return e.dispose(), a;
-    } };
-  }
-};
-h(x);
-function m(r, s) {
-  return !i(r) && !i(s) ? g(r, s) : A().runKernel("Mul16", { a: r, b: s });
-}
-export {
-  m as mul16
-};

package/dist/ops/mulDrop.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function mulDrop(a: Tensor, b: Tensor, dropoutRate: number, seed: number): Tensor;

package/dist/ops/mulDrop.js

@@ -1,9 +0,0 @@
-import { h as e } from "../index-CUXkjxiT.js";
-import "./cpu/mulDropout.js";
-import "./webgl/mulDropout.js";
-function m(r, o, n, t) {
-  return e().runKernel("MulDropout", { a: r, b: o }, { dropoutRate: n, seed: t });
-}
-export {
-  m as mulDrop
-};

package/dist/ops/normRMS.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function normRMS(x: Tensor, gamma?: Tensor): Tensor;

package/dist/ops/normRMS.js

@@ -1,19 +0,0 @@
-import { h as s } from "../index-CUXkjxiT.js";
-import "./cpu/normRMS.js";
-import "./webgl/normRMS.js";
-import "./grads/normRMS.js";
-function R(e, r) {
-  if (e.rank === 4) {
-    const [t, p, ...i] = e.shape, h = [t * p, ...i], o = e.reshape(h), n = s().runKernel(
-      r ? "RMSNorm" : "RMSNormNoGamma",
-      r ? { x: o, gamma: r } : { x: o }
-    );
-    o.dispose();
-    const c = n.reshape(e.shape);
-    return n.dispose(), c;
-  }
-  return s().runKernel(r ? "RMSNorm" : "RMSNormNoGamma", r ? { x: e, gamma: r } : { x: e });
-}
-export {
-  R as normRMS
-};

package/dist/ops/pack16.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function pack16(x: Tensor, scaling?: number, padding?: number): Tensor;

package/dist/ops/pack16.js

@@ -1,5 +0,0 @@
-import "../index-CUXkjxiT.js";
-import { p as a } from "../pack16-bqltoUlR.js";
-export {
-  a as pack16
-};

package/dist/ops/qkv.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function qkv(x: Tensor, kernel: Tensor, heads: number, packed?: boolean): Tensor[];

package/dist/ops/qkv.js

@@ -1,10 +0,0 @@
-import { h as e } from "../index-CUXkjxiT.js";
-import "./cpu/qkv.js";
-import "./webgl/qkv.js";
-import "./grads/qkv.js";
-function u(r, n, o, t = !1) {
-  return e().runKernel("QKV", { x: r, kernel: n }, { heads: o, packed: t });
-}
-export {
-  u as qkv
-};

package/dist/ops/reshape16.d.ts

@@ -1,2 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-export declare function reshape16(x: Tensor, shape: number[]): Tensor;