@genai-fi/nanogpt 0.3.2 → 0.4.0

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 (98) hide show
  1. package/dist/Generator.js +22 -22
  2. package/dist/MLP-KHhikThU.js +83 -0
  3. package/dist/NanoGPTModel.d.ts +2 -3
  4. package/dist/NanoGPTModel.js +79 -79
  5. package/dist/TeachableLLM.js +16 -13
  6. package/dist/axis_util-DeydwOoC.js +69 -0
  7. package/dist/{concat-BIZS_td9.js → concat-DS_qH7MI.js} +5 -5
  8. package/dist/config.js +7 -8
  9. package/dist/{gather-BPGW8RsB.js → gather-BUmJIS8n.js} +1 -1
  10. package/dist/{index-pWA4_lUh.js → index-XjBAhiFO.js} +1272 -1174
  11. package/dist/layers/BaseLayer.d.ts +14 -2
  12. package/dist/layers/BaseLayer.js +9 -9
  13. package/dist/layers/CausalSelfAttention.d.ts +4 -8
  14. package/dist/layers/CausalSelfAttention.js +108 -82
  15. package/dist/layers/MLP.d.ts +2 -3
  16. package/dist/layers/MLP.js +5 -62
  17. package/dist/layers/RMSNorm.d.ts +2 -2
  18. package/dist/layers/RMSNorm.js +11 -11
  19. package/dist/layers/RoPECache.js +3 -3
  20. package/dist/layers/TiedEmbedding.js +7 -6
  21. package/dist/layers/TransformerBlock.d.ts +2 -6
  22. package/dist/layers/TransformerBlock.js +9 -12
  23. package/dist/{sum-C7Mgy9Bw.js → log_sum_exp-DJPkVZZn.js} +32 -54
  24. package/dist/main.js +22 -19
  25. package/dist/{mat_mul-D7_a4KJn.js → mat_mul-CKwFEV1Q.js} +1 -1
  26. package/dist/max-DJvEiCAJ.js +25 -0
  27. package/dist/moments-CrWRPcR3.js +53 -0
  28. package/dist/norm-BzY929B_.js +86 -0
  29. package/dist/{ones-Cog-G2ag.js → ones-BO01zpJG.js} +2 -2
  30. package/dist/ops/appendCache.js +1 -1
  31. package/dist/ops/attentionMask.js +1 -1
  32. package/dist/ops/cpu/appendCache.js +2 -2
  33. package/dist/ops/cpu/attentionMask.js +2 -2
  34. package/dist/ops/cpu/fusedSoftmax.d.ts +9 -0
  35. package/dist/ops/cpu/fusedSoftmax.js +23 -0
  36. package/dist/ops/cpu/gatherSub.js +3 -3
  37. package/dist/ops/cpu/mulDropout.d.ts +1 -0
  38. package/dist/ops/cpu/mulDropout.js +17 -0
  39. package/dist/ops/cpu/qkv.js +3 -3
  40. package/dist/ops/cpu/rope.js +5 -5
  41. package/dist/ops/cpu/scatterSub.js +27 -27
  42. package/dist/ops/fusedSoftmax.d.ts +2 -0
  43. package/dist/ops/fusedSoftmax.js +10 -0
  44. package/dist/ops/gatherSub.js +1 -1
  45. package/dist/ops/grads/attentionMask.js +1 -1
  46. package/dist/ops/grads/fusedSoftmax.d.ts +2 -0
  47. package/dist/ops/grads/fusedSoftmax.js +17 -0
  48. package/dist/ops/grads/qkv.js +1 -1
  49. package/dist/ops/grads/rope.js +1 -1
  50. package/dist/ops/mulDrop.d.ts +2 -0
  51. package/dist/ops/mulDrop.js +9 -0
  52. package/dist/ops/node/sparseCrossEntropy.js +1 -1
  53. package/dist/ops/qkv.js +1 -1
  54. package/dist/ops/scatterSub.js +1 -1
  55. package/dist/ops/webgl/appendCache.js +1 -1
  56. package/dist/ops/webgl/attentionMask.js +1 -1
  57. package/dist/ops/webgl/fusedSoftmax.d.ts +11 -0
  58. package/dist/ops/webgl/fusedSoftmax.js +3930 -0
  59. package/dist/ops/webgl/gatherSub.js +1 -1
  60. package/dist/ops/webgl/mulDropout.d.ts +1 -0
  61. package/dist/ops/webgl/mulDropout.js +41 -0
  62. package/dist/ops/webgl/qkv.js +1 -1
  63. package/dist/ops/webgl/rope.js +1 -1
  64. package/dist/ops/webgl/scatterSub.js +1 -1
  65. package/dist/{random_width-oeUIlUZj.js → random_width-CMHmdbSu.js} +4212 -6630
  66. package/dist/{range-CcDl05lo.js → range-DQMNzBWs.js} +1 -1
  67. package/dist/{reshape-C8CR_Bad.js → reshape-DFzh97Sc.js} +1 -1
  68. package/dist/{sin-BJIrfnj7.js → sin-BYM-U4Ut.js} +1 -1
  69. package/dist/slice_util-CnVNPQI-.js +90 -0
  70. package/dist/softmax-4DOn6cPq.js +28 -0
  71. package/dist/{split-DZbvruEP.js → split-CkbeVdF8.js} +3 -3
  72. package/dist/{stack-BMm-efee.js → stack-DaIMO5iX.js} +1 -1
  73. package/dist/sum-C6u3xMi3.js +27 -0
  74. package/dist/{tensor-DJVbYhh1.js → tensor-Cu1fU7H7.js} +1 -1
  75. package/dist/{tensor2d-ZuQSh2D-.js → tensor2d-D0CKdG6B.js} +1 -1
  76. package/dist/tfjs_backend-Bzl2SrRo.js +2460 -0
  77. package/dist/training/AdamExt.js +1 -1
  78. package/dist/training/DatasetBuilder.js +3 -3
  79. package/dist/training/FullTrainer.js +1 -1
  80. package/dist/training/Trainer.js +13 -12
  81. package/dist/training/sparseCrossEntropy.js +12 -11
  82. package/dist/utilities/dummy.js +8 -8
  83. package/dist/utilities/generate.js +11 -11
  84. package/dist/utilities/load.js +1 -1
  85. package/dist/utilities/profile.js +1 -1
  86. package/dist/utilities/weights.js +2 -2
  87. package/dist/{variable-Dl_ub3pk.js → variable-BS4AKqNU.js} +1 -1
  88. package/dist/{zeros-CCy9C3uU.js → zeros-CmJFiC84.js} +1 -1
  89. package/package.json +1 -1
  90. package/dist/exports_layers-tbTBcwMM.js +0 -25
  91. package/dist/layers/LayerNorm.d.ts +0 -13
  92. package/dist/layers/LayerNorm.js +0 -33
  93. package/dist/moments-DfcpfwKi.js +0 -132
  94. package/dist/softmax-Be_lsqUc.js +0 -105
  95. package/dist/training/LayerTrainer.d.ts +0 -29
  96. package/dist/training/LayerTrainer.js +0 -95
  97. package/dist/training/lwSchedule.d.ts +0 -7
  98. package/dist/training/lwSchedule.js +0 -162
package/dist/config.js CHANGED
@@ -1,20 +1,19 @@
1
1
  const e = {
2
- vocabSize: 50304,
3
- // GPT-2 vocab size
4
- blockSize: 1024,
2
+ vocabSize: 2e3,
3
+ blockSize: 128,
5
4
  // Maximum sequence length
6
- nLayer: 12,
5
+ nLayer: 6,
7
6
  // Number of transformer layers
8
- nHead: 12,
7
+ nHead: 4,
9
8
  // Number of attention heads
10
- nEmbed: 768,
9
+ nEmbed: 256,
11
10
  // Embedding dimension
12
- dropout: 0,
11
+ dropout: 0.1,
13
12
  // Dropout probability
14
13
  biasInLinear: !1,
15
14
  biasInLayerNorm: !1,
16
15
  mlpFactor: 4,
17
- useRope: !1
16
+ useRope: !0
18
17
  // Use Rotary Position Embeddings
19
18
  };
20
19
  export {
@@ -1,4 +1,4 @@
1
- import { o as h, h as t, E as g, G as p } from "./index-pWA4_lUh.js";
1
+ import { o as h, h as t, E as g, G as p } from "./index-XjBAhiFO.js";
2
2
  /**
3
3
  * @license
4
4
  * Copyright 2018 Google LLC. All Rights Reserved.