@genai-fi/nanogpt 0.4.0 → 0.4.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (87) hide show
  1. package/dist/Generator.js +3 -3
  2. package/dist/NanoGPTModel.js +83 -70
  3. package/dist/TeachableLLM.js +1 -1
  4. package/dist/{random_width-CMHmdbSu.js → TiedEmbedding-CnJ1bx4q.js} +760 -719
  5. package/dist/{axis_util-DeydwOoC.js → axis_util-BgTGy5w8.js} +1 -1
  6. package/dist/{concat-DS_qH7MI.js → concat-CuRsVY-K.js} +1 -1
  7. package/dist/dropout-DfDdklfL.js +193 -0
  8. package/dist/{gather-BUmJIS8n.js → gather-ZYRWhmXR.js} +1 -1
  9. package/dist/gelu-CnCt17Lk.js +26 -0
  10. package/dist/{index-XjBAhiFO.js → index-C4JCoBvj.js} +61 -61
  11. package/dist/kernel_funcs_utils-CAd1h9X1.js +388 -0
  12. package/dist/layers/CausalSelfAttention.js +73 -72
  13. package/dist/layers/MLP.d.ts +3 -1
  14. package/dist/layers/MLP.js +93 -5
  15. package/dist/layers/RMSNorm.js +3 -3
  16. package/dist/layers/RoPECache.js +3 -3
  17. package/dist/layers/TiedEmbedding.js +6 -46
  18. package/dist/layers/TransformerBlock.js +2 -2
  19. package/dist/{log_sum_exp-DJPkVZZn.js → log_sum_exp-BswFnwOb.js} +5 -5
  20. package/dist/main.js +1 -1
  21. package/dist/{mat_mul-CKwFEV1Q.js → mat_mul-415y5Qn2.js} +1 -1
  22. package/dist/{max-DJvEiCAJ.js → max-CP_9O2Yd.js} +1 -1
  23. package/dist/{moments-CrWRPcR3.js → moments-CjeIaVdp.js} +3 -3
  24. package/dist/{norm-BzY929B_.js → norm-CZM380I3.js} +5 -5
  25. package/dist/{ones-BO01zpJG.js → ones-Bf3YR48P.js} +2 -2
  26. package/dist/ops/appendCache.js +1 -1
  27. package/dist/ops/attentionMask.d.ts +1 -1
  28. package/dist/ops/attentionMask.js +4 -4
  29. package/dist/ops/cpu/appendCache.js +2 -2
  30. package/dist/ops/cpu/attentionMask.js +13 -9
  31. package/dist/ops/cpu/fusedSoftmax.js +2 -2
  32. package/dist/ops/cpu/gatherSub.js +3 -3
  33. package/dist/ops/cpu/gelu.d.ts +1 -0
  34. package/dist/ops/cpu/gelu.js +40 -0
  35. package/dist/ops/cpu/mulDropout.js +1 -1
  36. package/dist/ops/cpu/qkv.js +3 -3
  37. package/dist/ops/cpu/rope.js +5 -5
  38. package/dist/ops/cpu/scatterSub.js +4 -4
  39. package/dist/ops/fusedSoftmax.js +1 -1
  40. package/dist/ops/gatherSub.js +1 -1
  41. package/dist/ops/gelu.d.ts +3 -0
  42. package/dist/ops/gelu.js +8 -0
  43. package/dist/ops/grads/attentionMask.js +1 -1
  44. package/dist/ops/grads/fusedSoftmax.js +2 -2
  45. package/dist/ops/grads/gelu.d.ts +2 -0
  46. package/dist/ops/grads/gelu.js +5 -0
  47. package/dist/ops/grads/qkv.js +1 -1
  48. package/dist/ops/grads/rope.js +1 -1
  49. package/dist/ops/mulDrop.js +1 -1
  50. package/dist/ops/node/sparseCrossEntropy.js +1 -1
  51. package/dist/ops/qkv.js +1 -1
  52. package/dist/ops/scatterSub.js +1 -1
  53. package/dist/ops/webgl/appendCache.js +1 -1
  54. package/dist/ops/webgl/attentionMask.js +19 -18
  55. package/dist/ops/webgl/fusedSoftmax.js +489 -788
  56. package/dist/ops/webgl/gatherSub.js +1 -1
  57. package/dist/ops/webgl/gelu.d.ts +2 -0
  58. package/dist/ops/webgl/gelu.js +50 -0
  59. package/dist/ops/webgl/mulDropout.js +1 -1
  60. package/dist/ops/webgl/qkv.js +1 -1
  61. package/dist/ops/webgl/rope.js +1 -1
  62. package/dist/ops/webgl/scatterSub.js +1 -1
  63. package/dist/{range-DQMNzBWs.js → range-9AzeApCc.js} +1 -1
  64. package/dist/{reshape-DFzh97Sc.js → reshape-Boe4DuIO.js} +1 -1
  65. package/dist/{sin-BYM-U4Ut.js → sin-KmhiDuMa.js} +1 -1
  66. package/dist/{slice_util-CnVNPQI-.js → slice_util-19zDNNSn.js} +2 -2
  67. package/dist/{softmax-4DOn6cPq.js → softmax-Cujsg4ay.js} +1 -1
  68. package/dist/{split-CkbeVdF8.js → split-DbcNm1-i.js} +1 -1
  69. package/dist/{stack-DaIMO5iX.js → stack-D1YjmgKN.js} +1 -1
  70. package/dist/{sum-C6u3xMi3.js → sum-R28pucR5.js} +1 -1
  71. package/dist/{tensor-Cu1fU7H7.js → tensor-BVeHdl7V.js} +1 -1
  72. package/dist/{tensor2d-D0CKdG6B.js → tensor2d-DqFGNs_K.js} +1 -1
  73. package/dist/{tfjs_backend-Bzl2SrRo.js → tfjs_backend-Cug-PH75.js} +826 -1015
  74. package/dist/training/AdamExt.js +1 -1
  75. package/dist/training/DatasetBuilder.js +3 -3
  76. package/dist/training/FullTrainer.js +1 -1
  77. package/dist/training/Trainer.js +5 -5
  78. package/dist/training/sparseCrossEntropy.js +4 -4
  79. package/dist/utilities/dummy.js +2 -2
  80. package/dist/utilities/generate.js +3 -3
  81. package/dist/utilities/load.js +1 -1
  82. package/dist/utilities/profile.js +1 -1
  83. package/dist/utilities/weights.js +2 -2
  84. package/dist/{variable-BS4AKqNU.js → variable-LJT9Ld63.js} +1 -1
  85. package/dist/{zeros-CmJFiC84.js → zeros-dnQxFgAD.js} +1 -1
  86. package/package.json +1 -1
  87. package/dist/MLP-KHhikThU.js +0 -83
@@ -1,4 +1,4 @@
1
- import { r as l } from "../../index-XjBAhiFO.js";
1
+ import { r as l } from "../../index-C4JCoBvj.js";
2
2
  class u {
3
3
  variableNames = ["labels", "logits", "values"];
4
4
  outputShape;
@@ -0,0 +1,2 @@
1
+ import { KernelFunc } from '@tensorflow/tfjs-core';
2
+ export declare const gelu: KernelFunc;
@@ -0,0 +1,50 @@
1
+ import { r as a } from "../../index-C4JCoBvj.js";
2
+ import { u as s, C as x } from "../../kernel_funcs_utils-CAd1h9X1.js";
3
+ const t = 0.7978845608028654, r = 0.044715, c = x + `
4
+ float x3 = x * x * x;
5
+ float inner = x + ${r} * x3;
6
+ inner = ${t} * inner;
7
+ inner = tanh(inner);
8
+ inner = 0.5 * (1.0 + inner);
9
+ x = x * inner;
10
+ return x;
11
+ `, d = s({ opSnippet: c }), i = {
12
+ kernelName: "Gelu",
13
+ backendName: "webgl",
14
+ kernelFunc: d
15
+ };
16
+ a(i);
17
+ class f {
18
+ // Inputs: dy, x
19
+ variableNames = ["dy", "x"];
20
+ outputShape;
21
+ userCode;
22
+ constructor(n) {
23
+ this.outputShape = n, this.userCode = `
24
+ void main() {
25
+ float dy = getDyAtOutCoords();
26
+ float x = getXAtOutCoords();
27
+ float x2 = x * x;
28
+ float x3 = x2 * x;
29
+ float u = ${t} * (x + ${r} * x3);
30
+ float t = tanh(u);
31
+ float sech2 = 1.0 - t * t;
32
+ float du_dx = ${t} * (1.0 + 3.0 * ${r} * x2);
33
+ float dgelu = 0.5 * (1.0 + t) + 0.5 * x * sech2 * du_dx;
34
+ setOutput(dy * dgelu);
35
+ }`;
36
+ }
37
+ }
38
+ function g(e) {
39
+ const { dy: n, x: o } = e.inputs, u = e.backend, l = new f(o.shape);
40
+ return u.runWebGLProgram(l, [n, o], "float32");
41
+ }
42
+ const p = {
43
+ kernelName: "GeluGrad",
44
+ backendName: "webgl",
45
+ kernelFunc: g
46
+ };
47
+ a(p);
48
+ export {
49
+ d as gelu
50
+ };
@@ -1,4 +1,4 @@
1
- import { r as m } from "../../index-XjBAhiFO.js";
1
+ import { r as m } from "../../index-C4JCoBvj.js";
2
2
  class f {
3
3
  variableNames = ["a", "b"];
4
4
  outputShape;
@@ -1,4 +1,4 @@
1
- import { r as i } from "../../index-XjBAhiFO.js";
1
+ import { r as i } from "../../index-C4JCoBvj.js";
2
2
  class l {
3
3
  variableNames = ["x", "kernel"];
4
4
  outputShape;
@@ -1,4 +1,4 @@
1
- import { r as u } from "../../index-XjBAhiFO.js";
1
+ import { r as u } from "../../index-C4JCoBvj.js";
2
2
  class l {
3
3
  variableNames = ["x", "sin", "cos"];
4
4
  outputShape;
@@ -1,4 +1,4 @@
1
- import { r as i } from "../../index-XjBAhiFO.js";
1
+ import { r as i } from "../../index-C4JCoBvj.js";
2
2
  class u {
3
3
  variableNames = ["labels", "softmaxProbs", "dy"];
4
4
  outputShape;
@@ -1,4 +1,4 @@
1
- import { E as e, R as f } from "./index-XjBAhiFO.js";
1
+ import { E as e, R as f } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { o, h as t, E as a, q as p } from "./index-XjBAhiFO.js";
1
+ import { o, h as t, E as a, q as p } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { o, h as t, E as c, Q as a, T as e } from "./index-XjBAhiFO.js";
1
+ import { o, h as t, E as c, Q as a, T as e } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -1,5 +1,5 @@
1
- import { o as u, h as p, k as g, w as m, E as w, a1 as x, j as i } from "./index-XjBAhiFO.js";
2
- import { r as y } from "./reshape-DFzh97Sc.js";
1
+ import { o as u, h as p, k as g, w as m, E as w, a1 as x, j as i } from "./index-C4JCoBvj.js";
2
+ import { r as y } from "./reshape-Boe4DuIO.js";
3
3
  /**
4
4
  * @license
5
5
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { o as r, h as f, E as e, S as i } from "./index-XjBAhiFO.js";
1
+ import { o as r, h as f, E as e, S as i } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { o as p, h as i, E as a, u as c } from "./index-XjBAhiFO.js";
1
+ import { o as p, h as i, E as a, u as c } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { o as e, i as c, j as n, E as i, P as k } from "./index-XjBAhiFO.js";
1
+ import { o as e, i as c, j as n, E as i, P as k } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { o as e, h as u, x as c, E as l, y as m } from "./index-XjBAhiFO.js";
1
+ import { o as e, h as u, x as c, E as l, y as m } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { J as t, K as a } from "./index-XjBAhiFO.js";
1
+ import { J as t, K as a } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -1,4 +1,4 @@
1
- import { I as t, J as s, K as a } from "./index-XjBAhiFO.js";
1
+ import { I as t, J as s, K as a } from "./index-C4JCoBvj.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.