@genai-fi/nanogpt 0.19.0 → 0.20.0

package/dist/mod-CSdCpRjf.js

@@ -1,11 +0,0 @@
-import { o as r, q as s, B as m, E as c, M as d } from "./index-CUXkjxiT.js";
-function p(t, e) {
-  let o = s(t, "a", "mod"), a = s(e, "b", "mod");
-  [o, a] = m(o, a);
-  const n = { a: o, b: a };
-  return c.runKernel(d, n);
-}
-const b = /* @__PURE__ */ r({ mod_: p });
-export {
-  b as m
-};

package/dist/models/NanoGPTV1.d.ts

@@ -1,16 +0,0 @@
-import { GPTConfigV1 } from './config';
-import { Tensor } from '@tensorflow/tfjs-core';
-import { default as Model, ModelForwardAttributes } from './model';
-export default class NanoGPTV1 extends Model<ModelForwardAttributes, GPTConfigV1> {
-    private wte;
-    private wpe?;
-    private blocks;
-    private lnF;
-    private ropeCache?;
-    constructor(config?: Partial<GPTConfigV1>);
-    getClassName(): string;
-    private inputPhase;
-    forward(attrs: ModelForwardAttributes, idx: Tensor): Tensor;
-    project(embeddings: Tensor): Tensor;
-    dispose(): void;
-}

package/dist/models/NanoGPTV1.js

@@ -1,99 +0,0 @@
-import u from "../layers/TransformerBlock.js";
-import f from "../layers/TiedEmbedding.js";
-import g from "../layers/RoPECache.js";
-import b from "../layers/RMSNorm.js";
-import { t as r, k as d } from "../index-CUXkjxiT.js";
-import k from "./model.js";
-import w from "../layers/PositionEmbedding.js";
-import { packingSupported as E } from "../utilities/packed.js";
-import { p as _, u as C } from "../pack16-bqltoUlR.js";
-const P = {
-  modelType: "GenAI_NanoGPT_v1",
-  vocabSize: 2e3,
-  blockSize: 128,
-  // Maximum sequence length
-  nLayer: 6,
-  // Number of transformer layers
-  nHead: 4,
-  // Number of attention heads
-  nEmbed: 256,
-  // Embedding dimension
-  mlpFactor: 4,
-  useRope: !0
-};
-class S extends k {
-  wte;
-  // Token embeddings
-  wpe;
-  // Position embeddings
-  // private drop: layers.Layer; // Dropout
-  blocks;
-  lnF;
-  // Final layer norm
-  ropeCache;
-  constructor(e = {}) {
-    super({ ...P, ...e });
-    const t = {
-      activation: "gelu",
-      hiddenFactor: this.config.mlpFactor,
-      useGamma: !0,
-      useQKNorm: !1
-    };
-    this.wte = new f(this.config, "token_embedding", this), this.config.useRope === !1 ? this.wpe = new w(this.config, "positional_embedding", this) : this.ropeCache = new g(this.config), this.blocks = [];
-    for (let o = 0; o < this.config.nLayer; o++)
-      this.blocks.push(new u(o, this.config, t, this));
-    this.lnF = new b(this.config, t, "final_rms_norm", this);
-  }
-  getClassName() {
-    return "GenAI_NanoGPT_v1";
-  }
-  inputPhase(e, t) {
-    return r(() => {
-      const o = this.wte.embed(e);
-      if (this.config.useRope === !1) {
-        const i = this.wpe.call(t, o);
-        if (Array.isArray(i))
-          throw new Error("PositionEmbedding output should not be an array");
-        return i;
-      }
-      return o;
-    });
-  }
-  forward(e, t) {
-    return this.validateInput(t), e.ropeCache = this.ropeCache, e.outputEmbeddings && (e.embeddings = []), r(() => {
-      this.startMemory();
-      let o = this.inputPhase(t, e);
-      if (e.cache && e.cache.length !== this.blocks.length)
-        throw console.error("Cache", e.cache), new Error(
-          `Cache length ${e.cache.length} does not match number of blocks ${this.blocks.length}`
-        );
-      const i = e.mixedPrecision === !0 && E();
-      let s = i ? _(o) : o;
-      i && o !== s && o.dispose();
-      for (let n = 0; n < this.blocks.length; n++) {
-        const p = this.blocks[n], a = Math.random() * 1e9, l = {
-          ...e,
-          seed: a,
-          pastKV: e.cache ? e.cache[n] : void 0,
-          mixedPrecision: i
-        }, m = e.checkpointing && e.training ? p.callCheckpoint(l, s) : p.call(l, s);
-        e.outputEmbeddings ? (d(s), e.embeddings.push({ name: `block_output_${n}`, tensor: s })) : s.dispose(), s = m;
-      }
-      if (o = this.lnF.call({ ...e, mixedPrecision: i }, s), s.dispose(), e.skipLogits)
-        return this.endMemory("Forward"), o;
-      const c = this.wte.project(o);
-      e.outputEmbeddings ? (d(o), e.embeddings.push({ name: "final_norm_output", tensor: o })) : o.dispose();
-      const h = i ? C(c) : c;
-      return i && c !== h && c.dispose(), h;
-    });
-  }
-  project(e) {
-    return r(() => this.wte.project(e));
-  }
-  dispose() {
-    this.weightStore.dispose(), this.wte.dispose(), this.wpe && this.wpe.dispose(), this.blocks.forEach((e) => e.dispose()), this.lnF.dispose();
-  }
-}
-export {
-  S as default
-};

package/dist/models/NanoGPTV2.d.ts

@@ -1,16 +0,0 @@
-import { GPTConfigV2 } from './config';
-import { Tensor } from '@tensorflow/tfjs-core';
-import { default as Model, ModelForwardAttributes } from './model';
-export default class NanoGPTV2 extends Model<ModelForwardAttributes, GPTConfigV2> {
-    private wte;
-    private wpe?;
-    private blocks;
-    private lnF;
-    private ropeCache?;
-    constructor(config?: Partial<GPTConfigV2>);
-    getClassName(): string;
-    private inputPhase;
-    forward(attrs: ModelForwardAttributes, idx: Tensor): Tensor;
-    project(embeddings: Tensor): Tensor;
-    dispose(): void;
-}

package/dist/models/NanoGPTV2.js

@@ -1,90 +0,0 @@
-import u from "../layers/TransformerBlock.js";
-import f from "../layers/TiedEmbedding.js";
-import g from "../layers/RoPECache.js";
-import b from "../layers/RMSNorm.js";
-import { t as r, k as d } from "../index-CUXkjxiT.js";
-import k from "./model.js";
-import { packingSupported as w } from "../utilities/packed.js";
-import { p as C, u as _ } from "../pack16-bqltoUlR.js";
-const E = {
-  modelType: "GenAI_NanoGPT_v2",
-  vocabSize: 2e3,
-  blockSize: 128,
-  // Maximum sequence length
-  nLayer: 6,
-  // Number of transformer layers
-  nHead: 4,
-  // Number of attention heads
-  nEmbed: 256,
-  // Embedding dimension
-  mlpFactor: 4
-};
-class S extends k {
-  wte;
-  // Token embeddings
-  wpe;
-  // Position embeddings
-  // private drop: layers.Layer; // Dropout
-  blocks;
-  lnF;
-  // Final layer norm
-  ropeCache;
-  constructor(e = {}) {
-    super({ ...E, ...e });
-    const s = {
-      activation: "relu2",
-      hiddenFactor: this.config.mlpFactor,
-      useGamma: !1,
-      useQKNorm: !0
-    };
-    this.wte = new f(this.config, "token_embedding", this), this.ropeCache = new g(this.config), this.blocks = [];
-    for (let o = 0; o < this.config.nLayer; o++)
-      this.blocks.push(new u(o, this.config, s, this));
-    this.lnF = new b(this.config, s, "final_rms_norm", this);
-  }
-  getClassName() {
-    return "GenAI_NanoGPT_v2";
-  }
-  inputPhase(e) {
-    return r(() => this.wte.embed(e));
-  }
-  forward(e, s) {
-    return this.validateInput(s), e.ropeCache = this.ropeCache, e.outputEmbeddings && (e.embeddings = []), r(() => {
-      this.startMemory();
-      let o = this.inputPhase(s);
-      if (e.cache && e.cache.length !== this.blocks.length)
-        throw console.error("Cache", e.cache), new Error(
-          `Cache length ${e.cache.length} does not match number of blocks ${this.blocks.length}`
-        );
-      const t = e.mixedPrecision === !0 && w();
-      let i = t ? C(o) : o;
-      t && o !== i && o.dispose();
-      for (let n = 0; n < this.blocks.length; n++) {
-        if (e.layerDrop && Math.random() < e.layerDrop * (n / this.blocks.length))
-          continue;
-        const p = this.blocks[n], m = Math.random() * 1e9, l = {
-          ...e,
-          seed: m,
-          pastKV: e.cache ? e.cache[n] : void 0,
-          mixedPrecision: t
-        }, a = e.checkpointing && e.training ? p.callCheckpoint(l, i) : p.call(l, i);
-        e.outputEmbeddings ? (d(i), e.embeddings.push({ name: `block_output_${n}`, tensor: i })) : i.dispose(), i = a;
-      }
-      if (o = this.lnF.call({ ...e, mixedPrecision: t }, i), i.dispose(), e.skipLogits)
-        return this.endMemory("Forward"), o;
-      const c = this.wte.project(o);
-      e.outputEmbeddings ? (d(o), e.embeddings.push({ name: "final_norm_output", tensor: o })) : o.dispose();
-      const h = t ? _(c) : c;
-      return t && c !== h && c.dispose(), h;
-    });
-  }
-  project(e) {
-    return r(() => this.wte.project(e));
-  }
-  dispose() {
-    this.weightStore.dispose(), this.wte.dispose(), this.wpe && this.wpe.dispose(), this.blocks.forEach((e) => e.dispose()), this.lnF.dispose();
-  }
-}
-export {
-  S as default
-};

package/dist/models/config.d.ts

@@ -1,27 +0,0 @@
-export interface LoRAConfig {
-    rank: number;
-    alpha: number;
-    variables: string[];
-}
-export interface GPTConfigBase {
-    modelType?: string;
-    vocabSize: number;
-    blockSize: number;
-    nLayer: number;
-    nHead: number;
-    nEmbed: number;
-    mlpFactor: number;
-    loraConfig?: Map<string, LoRAConfig>;
-    loraName?: string;
-}
-export interface GPTConfigV1 extends GPTConfigBase {
-    modelType: 'GenAI_NanoGPT_v1';
-    useRope: boolean;
-}
-export interface GPTConfigV2 extends GPTConfigBase {
-    modelType: 'GenAI_NanoGPT_v2';
-    windowSize?: string;
-}
-export type GPTConfig = GPTConfigV1 | GPTConfigV2;
-export declare const defaultConfig: GPTConfig;
-export declare function validateConfig(config: unknown): asserts config is GPTConfig;

package/dist/models/config.js

@@ -1,50 +0,0 @@
-const t = {
-  modelType: "GenAI_NanoGPT_v2",
-  vocabSize: 2e3,
-  blockSize: 128,
-  // Maximum sequence length
-  nLayer: 6,
-  // Number of transformer layers
-  nHead: 4,
-  // Number of attention heads
-  nEmbed: 256,
-  // Embedding dimension
-  mlpFactor: 4
-};
-function n(e, a) {
-  if (typeof e[a] != "number" || Number.isNaN(e[a]))
-    throw new Error(`Invalid config: "${a}" must be a number.`);
-}
-function l(e) {
-  const a = (o) => typeof o == "object" && o !== null && !Array.isArray(o);
-  if (!a(e))
-    throw new Error("Invalid config: expected an object.");
-  if (n(e, "vocabSize"), n(e, "blockSize"), n(e, "nLayer"), n(e, "nHead"), n(e, "nEmbed"), n(e, "mlpFactor"), e.loraConfig !== void 0) {
-    if (!a(e.loraConfig))
-      throw new Error('Invalid config: "loraConfig" must be an object.');
-    const o = Object.values(e.loraConfig);
-    if (!o.every((r) => a(r)))
-      throw new Error('Invalid config: each entry in "loraConfig" must be an object.');
-    if (!o.every((r) => "rank" in r && "alpha" in r && "variables" in r))
-      throw new Error('Invalid config: each LoRA config must have "rank", "alpha", and "variables" fields.');
-    o.forEach((r) => {
-      if (n(r, "rank"), n(r, "alpha"), !Array.isArray(r.variables) || !r.variables.every((i) => typeof i == "string"))
-        throw new Error('Invalid config: "variables" must be a string array.');
-    });
-  }
-  if (e.modelType === "GenAI_NanoGPT_v1") {
-    if (typeof e.useRope != "boolean")
-      throw new Error('Invalid config: "useRope" must be a boolean for GenAI_NanoGPT_v1.');
-    return;
-  }
-  if (e.modelType === "GenAI_NanoGPT_v2") {
-    if (e.windowSize !== void 0 && typeof e.windowSize != "string")
-      throw new Error('Invalid config: "windowSize" must be a string for GenAI_NanoGPT_v2.');
-    return;
-  }
-  throw new Error('Invalid config: "modelType" must be "GenAI_NanoGPT_v1" or "GenAI_NanoGPT_v2".');
-}
-export {
-  t as defaultConfig,
-  l as validateConfig
-};

package/dist/models/factory.d.ts

@@ -1,3 +0,0 @@
-import { GPTConfig } from './config';
-import { default as Model, ModelForwardAttributes } from './model';
-export default function createModelInstance(config: GPTConfig): Model<ModelForwardAttributes, GPTConfig>;

package/dist/models/factory.js

@@ -1,16 +0,0 @@
-import o from "./NanoGPTV1.js";
-import r from "./NanoGPTV2.js";
-function p(e) {
-  const n = e.modelType || "GenAI_NanoGPT_v1";
-  switch (n) {
-    case "GenAI_NanoGPT_v1":
-      return new o(e);
-    case "GenAI_NanoGPT_v2":
-      return new r(e);
-    default:
-      throw new Error(`Unsupported model type: ${n}`);
-  }
-}
-export {
-  p as default
-};

package/dist/models/model.d.ts

@@ -1,44 +0,0 @@
-import { Tensor } from '@tensorflow/tfjs-core';
-import { ForwardAttributes, default as BaseLayer } from '../layers/BaseLayer';
-import { AttentionScores, KVCache } from '../layers/CausalSelfAttention';
-import { TransformersMetadata } from '../loader/types';
-import { GPTConfig, LoRAConfig } from './config';
-import { default as LoRA } from '../layers/LoRA';
-export interface ModelForwardAttributes extends ForwardAttributes {
-    cache?: KVCache[];
-    attentionScores?: AttentionScores;
-    seed?: number;
-    skipLogits?: boolean;
-    ropePositionOffset?: number;
-}
-export interface TrainingState {
-    steps: number;
-    learningRate: number;
-    batchSize: number;
-    loss: number;
-    tokensProcessed: number;
-    duration: number;
-}
-export default abstract class Model<T extends ModelForwardAttributes, C extends GPTConfig = GPTConfig> extends BaseLayer<T, C> {
-    lossScaling: number;
-    trainingState: TrainingState | null;
-    metaData: TransformersMetadata;
-    private loraLayer?;
-    private loraMap;
-    constructor(config: C);
-    createLoRA(name: string, loraConfig: LoRAConfig): void;
-    deleteLoRA(name: string): void;
-    renameLoRA(oldName: string, newName: string): void;
-    mergeLoRA(name: string): void;
-    attachLoRA(name: string): void;
-    detachLoRA(): void;
-    hasLoRA(name?: string): boolean;
-    listLoRAs(): string[];
-    get lora(): LoRA | null;
-    abstract getClassName(): string;
-    abstract forward(attrs: T, idx: Tensor): Tensor;
-    abstract project(embeddings: Tensor): Tensor;
-    abstract dispose(): void;
-    getNumParams(): number;
-    protected validateInput(idx: Tensor): void;
-}

package/dist/models/model.js

@@ -1,134 +0,0 @@
-import i from "../layers/BaseLayer.js";
-import "../index-CUXkjxiT.js";
-import "../random_width-DN5ZtQkM.js";
-import "../zeros_like-CWjDdwr-.js";
-import "../Generator.js";
-import "../index-Cp39cXWe.js";
-import "../dataset-CGGp1z9P.js";
-import "../ops/cpu/attentionMask.js";
-import "../ops/webgl/attentionMask.js";
-import "../ops/grads/attentionMask.js";
-import "../ops/cpu/rope.js";
-import "../ops/webgl/rope.js";
-import "../rope-DcrZM_e6.js";
-import "../ops/cpu/appendCache.js";
-import "../ops/webgl/appendCache.js";
-import "../ops/grads/softmax16.js";
-import "../matMul16-BcVC_E62.js";
-import "../ops/webgl/matMul16.js";
-import "../ops/cpu/matMul16.js";
-import "../pack16-bqltoUlR.js";
-import "../ops/transpose16.js";
-import "../ops/reshape16.js";
-import "../ops/cpu/qkv.js";
-import "../ops/webgl/qkv.js";
-import "../ops/grads/qkv.js";
-import "../ops/cpu/normRMS.js";
-import "../ops/webgl/normRMS.js";
-import "../ops/grads/normRMS.js";
-import "../ops/dropout16.js";
-import "../ops/webgl/dropout16.js";
-import "../ops/grads/add16.js";
-import "../jszip.min-BZhlzntC.js";
-import "../index-DvYrXKkX.js";
-import "../ops/cpu/adamAdjust.js";
-import "../ops/webgl/adamAdjust.js";
-import "../ops/cpu/adamMoments.js";
-import "../ops/webgl/adamMoments.js";
-import "../ops/cpu/gatherSub.js";
-import "../ops/webgl/gatherSub.js";
-import "../ops/cpu/scatterSub.js";
-import "../ops/webgl/scatterSub.js";
-import "../papaparse.min-C0cScC2i.js";
-import { estimateParameterCount as a } from "../utilities/parameters.js";
-import "../ops/cpu/matMulGelu.js";
-import "../matMulGelu-JNLZqKQp.js";
-import "../ops/grads/matMulGelu.js";
-import "../ops/cpu/gelu.js";
-import "../ops/webgl/gelu.js";
-import "../gelu-B220X1Go.js";
-import "../ops/webgl/log.js";
-import "../checks/normRMS.js";
-import "../checks/normRMSGrad.js";
-import e from "../layers/LoRA.js";
-class st extends i {
-  lossScaling = 128;
-  trainingState = null;
-  metaData = { version: 2, application: "@genai-fi/nanogpt" };
-  loraLayer;
-  loraMap = /* @__PURE__ */ new Map();
-  constructor(t) {
-    super(t), t.loraConfig && t.loraConfig.forEach((r, o) => {
-      this.createLoRA(o, r);
-    });
-  }
-  createLoRA(t, r) {
-    if (this.loraMap.has(t))
-      return;
-    const o = new e(t, this.weightStore, r.alpha, r.rank, r.variables);
-    this.loraMap.set(t, o), this.config.loraConfig = this.config.loraConfig || /* @__PURE__ */ new Map(), this.config.loraConfig.set(t, r);
-  }
-  deleteLoRA(t) {
-    const r = this.loraMap.get(t);
-    if (!r)
-      throw new Error(`No LoRA with name ${t} exists.`);
-    this.loraLayer === r && this.detachLoRA(), r.dispose(), this.loraMap.delete(t), this.config.loraConfig && this.config.loraConfig.delete(t);
-  }
-  renameLoRA(t, r) {
-    if (!this.loraMap.has(t))
-      throw new Error(`No LoRA with name ${t} exists.`);
-    if (this.loraMap.has(r))
-      throw new Error(`LoRA with name ${r} already exists.`);
-    const o = this.loraMap.get(t);
-    this.loraMap.set(r, o), this.loraMap.delete(t), this.config.loraConfig && (this.config.loraConfig.delete(t), this.config.loraConfig.set(r, {
-      rank: o.rank,
-      alpha: o.alpha,
-      variables: Array.from(o.variables)
-    }));
-  }
-  mergeLoRA(t) {
-    const r = this.loraMap.get(t);
-    if (!r)
-      throw new Error(`No LoRA with name ${t} exists.`);
-    r.merge(), this.deleteLoRA(t);
-  }
-  attachLoRA(t) {
-    if (this.loraLayer) {
-      if (this.loraLayer.name === t)
-        return;
-      this.detachLoRA();
-    }
-    const r = this.loraMap.get(t);
-    if (!r)
-      throw new Error(`No LoRA with name ${t} exists.`);
-    r.attach(), this.loraLayer = r, this.config.loraName = t;
-  }
-  detachLoRA() {
-    if (!this.loraLayer)
-      throw new Error("No LoRA layer is attached to this model.");
-    this.loraLayer.detach(), this.loraLayer = void 0, this.config.loraName = void 0;
-  }
-  hasLoRA(t) {
-    return t ? this.loraMap.has(t) : !!this.loraLayer;
-  }
-  listLoRAs() {
-    return Array.from(this.loraMap.keys());
-  }
-  get lora() {
-    return this.loraLayer || null;
-  }
-  getNumParams() {
-    return a(this.config);
-  }
-  validateInput(t) {
-    if (t.shape.length !== 2)
-      throw new Error(`Invalid input shape: expected [batch_size, sequence_length], got ${t.shape}`);
-    if (t.shape[1] > this.config.blockSize)
-      throw new Error(`Input sequence length ${t.shape[1]} isn't block size ${this.config.blockSize}`);
-    if (t.dtype !== "int32")
-      throw new Error(`Input tensor must be of type int32, got ${t.dtype}`);
-  }
-}
-export {
-  st as default
-};

package/dist/non_max_suppression_impl-B2W7YjZB.js

@@ -1,102 +0,0 @@
-function J(e, t, o) {
-  const n = k(e, t, o), s = n < 0 ? -(n + 1) : n;
-  e.splice(s, 0, t);
-}
-function k(e, t, o) {
-  return A(e, t, o || v);
-}
-function v(e, t) {
-  return e > t ? 1 : e < t ? -1 : 0;
-}
-function A(e, t, o) {
-  let n = 0, s = e.length, r = 0, u = !1;
-  for (; n < s; ) {
-    r = n + (s - n >>> 1);
-    const l = o(t, e[r]);
-    l > 0 ? n = r + 1 : (s = r, u = !l);
-  }
-  return u ? n : -n - 1;
-}
-function O(e, t, o, n, s) {
-  return y(
-    e,
-    t,
-    o,
-    n,
-    s,
-    0
-    /* softNmsSigma */
-  );
-}
-function P(e, t, o, n, s, r) {
-  return y(
-    e,
-    t,
-    o,
-    n,
-    s,
-    0,
-    !1,
-    r,
-    !0
-    /* returnValidOutputs */
-  );
-}
-function X(e, t, o, n, s, r) {
-  return y(
-    e,
-    t,
-    o,
-    n,
-    s,
-    r,
-    !0
-    /* returnScoresTensor */
-  );
-}
-function y(e, t, o, n, s, r, u = !1, l = !1, h = !1) {
-  const c = [];
-  for (let i = 0; i < t.length; i++)
-    t[i] > s && c.push({ score: t[i], boxIndex: i, suppressBeginIndex: 0 });
-  c.sort(C);
-  const d = r > 0 ? -0.5 / r : 0, a = [], f = [];
-  for (; a.length < o && c.length > 0; ) {
-    const i = c.pop(), { score: M, boxIndex: I, suppressBeginIndex: g } = i;
-    if (M < s)
-      break;
-    let S = !1;
-    for (let b = a.length - 1; b >= g; --b) {
-      const w = B(e, I, a[b]);
-      if (w >= n) {
-        S = !0;
-        break;
-      }
-      if (i.score = i.score * V(n, d, w), i.score <= s)
-        break;
-    }
-    i.suppressBeginIndex = a.length, S || (i.score === M ? (a.push(I), f.push(i.score)) : i.score > s && J(c, i, C));
-  }
-  const x = a.length, p = o - x;
-  l && p > 0 && (a.push(...new Array(p).fill(0)), f.push(...new Array(p).fill(0)));
-  const m = { selectedIndices: a };
-  return u && (m.selectedScores = f), h && (m.validOutputs = x), m;
-}
-function B(e, t, o) {
-  const n = e.subarray(t * 4, t * 4 + 4), s = e.subarray(o * 4, o * 4 + 4), r = Math.min(n[0], n[2]), u = Math.min(n[1], n[3]), l = Math.max(n[0], n[2]), h = Math.max(n[1], n[3]), c = Math.min(s[0], s[2]), d = Math.min(s[1], s[3]), a = Math.max(s[0], s[2]), f = Math.max(s[1], s[3]), x = (l - r) * (h - u), p = (a - c) * (f - d);
-  if (x <= 0 || p <= 0)
-    return 0;
-  const m = Math.max(r, c), i = Math.max(u, d), M = Math.min(l, a), I = Math.min(h, f), g = Math.max(M - m, 0) * Math.max(I - i, 0);
-  return g / (x + p - g);
-}
-function V(e, t, o) {
-  const n = Math.exp(t * o * o);
-  return o <= e ? n : 0;
-}
-function C(e, t) {
-  return e.score - t.score || e.score === t.score && t.boxIndex - e.boxIndex;
-}
-export {
-  X as a,
-  P as b,
-  O as n
-};

package/dist/not_equal-hurPF26l.js

@@ -1,64 +0,0 @@
-import { o as c, q as e, _ as h, E as u, aL as b, aM as E, aN as _, aO as g, ah as S, a1 as w, j as K, aP as q, B as M, aQ as T } from "./index-CUXkjxiT.js";
-import { b as d } from "./broadcast_to-eS93CCN_.js";
-import { e as N } from "./axis_util-GTVlo58H.js";
-import { r as f } from "./reshape-BDOuCSNW.js";
-import { s as A } from "./sum-BpiwSWvg.js";
-function B(s, o, n) {
-  const a = e(o, "a", "where"), t = e(n, "b", "where"), r = e(s, "condition", "where", "bool"), p = h(h(r.shape, a.shape), t.shape), l = d(r, p), m = d(a, p), x = d(t, p), i = {
-    condition: l,
-    t: m,
-    e: x
-  };
-  return u.runKernel(b, i);
-}
-const H = /* @__PURE__ */ c({ where_: B });
-function G(s, o = null, n = !1) {
-  const t = { x: e(s, "x", "max") }, r = { reductionIndices: o, keepDims: n };
-  return u.runKernel(E, t, r);
-}
-const I = /* @__PURE__ */ c({ max_: G });
-function L(s) {
-  const n = { x: e(s, "x", "exp") };
-  return u.runKernel(_, n);
-}
-const P = /* @__PURE__ */ c({ exp_: L });
-function j(s) {
-  const n = { x: e(s, "x", "log", "float32") };
-  return u.runKernel(g, n);
-}
-const k = /* @__PURE__ */ c({ log_: j });
-function v(s, o = null, n = !1) {
-  const a = e(s, "x", "logSumExp"), t = S(o, a.shape), r = I(
-    a,
-    t,
-    !0
-    /* keepDims */
-  ), p = w(a, r), l = P(p), m = A(l, t), x = k(m), i = K(f(r, x.shape), x);
-  if (n) {
-    const $ = N(i.shape, t);
-    return f(i, $);
-  }
-  return i;
-}
-const J = /* @__PURE__ */ c({ logSumExp_: v });
-function y(s, o = null, n = !1) {
-  const t = { x: e(s, "x", "mean") }, r = { axis: o, keepDims: n };
-  return u.runKernel(q, t, r);
-}
-const R = /* @__PURE__ */ c({ mean_: y });
-function C(s, o) {
-  let n = e(s, "a", "notEqual", "string_or_numeric"), a = e(o, "b", "notEqual", "string_or_numeric");
-  [n, a] = M(n, a), h(n.shape, a.shape);
-  const t = { a: n, b: a };
-  return u.runKernel(T, t);
-}
-const U = /* @__PURE__ */ c({ notEqual_: C });
-export {
-  R as a,
-  k as b,
-  P as e,
-  J as l,
-  I as m,
-  U as n,
-  H as w
-};