@genai-fi/nanogpt 0.6.3 → 0.7.1

package/dist/training/sparseCrossEntropy.js

@@ -1,22 +1,22 @@
 import { gatherSub as x } from "../ops/gatherSub.js";
 import { scatterSub as L } from "../ops/scatterSub.js";
-import { l as C, t as u, z as E, b as G } from "../index-BAzbokzv.js";
-import { s as y } from "../softmax-fsdtf6JC.js";
-import { m as z } from "../max-DtlIuVeW.js";
-import { l as v } from "../log_sum_exp-YEo2h3gb.js";
+import { y, t as u, z as C, c as E } from "../index-UdZhlibC.js";
+import { s as G } from "../softmax-JMEIUo2J.js";
+import { m as z } from "../max-ByjEGoFx.js";
+import { l as v } from "../log_sum_exp-BnmCkHWl.js";
 function k(t, s) {
   return u(() => {
-    const n = t.shape[t.shape.length - 1], c = t.shape.slice(0, -1).reduce((o, e) => o * e, 1), h = t.shape.length > 2 ? t.reshape([c, n]) : t, p = s.shape.length > 1 ? s.reshape([c]).cast("int32") : s.cast("int32"), r = z(h, -1, !0), a = G(h, r), m = v(a, -1);
+    const n = t.shape[t.shape.length - 1], c = t.shape.slice(0, -1).reduce((o, e) => o * e, 1), h = t.shape.length > 2 ? t.reshape([c, n]) : t, p = s.shape.length > 1 ? s.reshape([c]).cast("int32") : s.cast("int32"), r = z(h, -1, !0), a = E(h, r), m = v(a, -1);
     return x(m, p, a);
   });
 }
 function A() {
-  return C(
+  return y(
     // @ts-expect-error Invalid params
     (s, n, d) => {
       const c = s.shape[s.shape.length - 1], p = s.shape.slice(0, -1).reduce((o, e) => o * e, 1), r = s.reshape([p, c]), a = n.reshape([p]).cast("int32"), m = k(r, a);
       return d([r, a]), r.dispose(), a.dispose(), { value: m, gradFunc: (o, e) => u(() => {
-        const S = e[0], f = e[1], b = y(S), l = L(b, f, o), g = E(n);
+        const S = e[0], f = e[1], b = G(S), l = L(b, f, o), g = C(n);
         return [l.reshape(s.shape), g];
       }) };
     }

package/dist/utilities/arrayClose.d.ts

@@ -0,0 +1 @@
+export declare function arraysClose(a: unknown, b: unknown, epsilon?: number): boolean;

package/dist/utilities/arrayClose.js

@@ -0,0 +1,11 @@
+function f(r, e, n = 1e-5) {
+  if (Array.isArray(r) && Array.isArray(e)) {
+    if (r.length !== e.length) return !1;
+    for (let t = 0; t < r.length; ++t)
+      if (!f(r[t], e[t], n)) return !1;
+    return !0;
+  } else return typeof r == "number" && typeof e == "number" ? r === -1 / 0 && e === -1 / 0 ? !0 : Math.abs(r - e) < n : !1;
+}
+export {
+  f as arraysClose
+};

package/dist/utilities/dummy.js

@@ -1,35 +1,35 @@
-import { m as f, v as S, e as w } from "../index-BAzbokzv.js";
-import { z as i } from "../zeros-YJDE7oRb.js";
-async function P(s) {
+import { m as y, v as P, e as S } from "../index-UdZhlibC.js";
+import { z as i } from "../zeros-DnPT2nD4.js";
+async function w(s) {
   const t = i([1, s.config.gpt.blockSize], "int32"), [e, n] = s.forward({ training: !1 }, t);
   await e.data(), e.dispose(), n && n.dispose(), t.dispose();
 }
-async function v(s) {
-  const t = f(), e = t.numBytesInGPUAllocated ?? t.numBytes;
-  await P(s);
+async function k(s) {
+  const t = y(), e = t.numBytesInGPUAllocated ?? t.numBytesAllocatedInGPU ?? t.numBytes;
+  await w(s);
   const n = i([1, s.config.gpt.blockSize], "int32"), r = i([1, s.config.gpt.blockSize], "int32"), o = {
     perBatch: 0,
     tapeSize: 0,
     gradients: s.getNumParams() * 4
-  }, g = () => {
-    const [a, l] = s.forward({ training: !0 }, n, r), d = w().state.activeTape;
+  }, f = () => {
+    const [c, l] = s.forward({ training: !0 }, n, r), p = S().state.activeTape;
     let u = 0;
-    if (d)
-      for (const z of d)
+    if (p)
+      for (const z of p)
         u += z.saved?.reduce((B, I) => B + I.size * 4, 0) || 0;
-    return o.tapeSize = u, a.dispose(), l;
-  }, { value: c, grads: m } = S(g), p = f(), y = p.numBytesInGPUAllocated ?? p.numBytes;
-  o.perBatch = y - e - o.gradients, console.log("Dummy training memory requirements:", o), await c.data(), c.dispose();
-  for (const a in m)
-    m[a].dispose();
+    return o.tapeSize = u, c.dispose(), l;
+  }, { value: m, grads: d } = P(f), a = y(), g = a.numBytesInGPUAllocated ?? a.numBytesAllocatedInGPU ?? a.numBytes;
+  o.perBatch = g - e - o.gradients, console.log("Dummy training memory requirements:", o), await m.data(), m.dispose();
+  for (const c in d)
+    d[c].dispose();
   return n.dispose(), r.dispose(), o;
 }
-function A(s) {
+function v(s) {
   const t = i([1, s.config.gpt.blockSize], "int32"), [e, n] = s.forward({ training: !1 }, t);
   e.dispose(), n && n.dispose(), t.dispose();
 }
 export {
-  A as dummyPass,
-  P as dummyPassAsync,
-  v as dummyPassTrainAsync
+  v as dummyPass,
+  w as dummyPassAsync,
+  k as dummyPassTrainAsync
 };

package/dist/utilities/generate.js

@@ -1,23 +1,22 @@
-import { t as y } from "../index-BAzbokzv.js";
-import { t as x } from "../tensor2d-CPEkynbH.js";
-import { c as f } from "../concat-5aPGqw3Z.js";
-async function A(o, r, a, c, T) {
+import "../index-UdZhlibC.js";
+import { t as m } from "../tensor2d-CWHxHpLh.js";
+import { c as u } from "../concat-CbXTetof.js";
+async function v(o, r, a, c, f) {
   if (c <= 0)
     throw new Error("Length must be a positive integer");
   if (a.length === 0)
     throw new Error("Prompt cannot be an empty string");
-  const p = await o.tokenise([a], !0), s = r.config.gpt.useRope ? new Array(r.config.gpt.nLayer).fill(void 0) : void 0, d = y(() => {
-    let t = x(p, [1, p[0].length], "int32"), e = t;
-    for (let g = 0; g < c; g++) {
-      const { output: n } = r.generate(t, s, T), l = t, m = e;
-      e = f([e, n], 1), t = s ? n : f([t, n], 1), l.dispose(), m.dispose(), s || n.dispose();
-    }
-    return e;
-  }), w = await d.array();
-  d.dispose();
-  const i = w[0], u = i.indexOf(o.eosToken);
-  return u !== -1 && i.splice(u), await o.decode(i);
+  const p = await o.tokenise([a], !0), s = r.config.gpt.useRope ? new Array(r.config.gpt.nLayer).fill(void 0) : void 0;
+  let e = m(p, [1, p[0].length], "int32"), t = e;
+  for (let g = 0; g < c; g++) {
+    const { output: n } = await r.generate(e, s, f), T = e, l = t;
+    t = u([t, n], 1), e = s ? n : u([e, n], 1), T.dispose(), l.dispose(), s || n.dispose();
+  }
+  const w = await t.array();
+  t.dispose();
+  const i = w[0], d = i.indexOf(o.eosToken);
+  return d !== -1 && i.splice(d), await o.decode(i);
 }
 export {
-  A as generateText
+  v as generateText
 };

package/dist/utilities/multinomialCPU.d.ts

@@ -0,0 +1,2 @@
+import { Tensor2D } from '@tensorflow/tfjs-core';
+export default function multinomialCPU(probs: number[]): Tensor2D;

package/dist/utilities/multinomialCPU.js

@@ -0,0 +1,13 @@
+import "../index-UdZhlibC.js";
+import { t as e } from "../tensor2d-CWHxHpLh.js";
+function l(n) {
+  let r = 0;
+  const i = Math.random();
+  for (let t = 0; t < n.length; t++)
+    if (r += n[t], i < r)
+      return e([[t]], [1, 1], "int32");
+  return e([[n.length - 1]], [1, 1], "int32");
+}
+export {
+  l as default
+};

package/dist/utilities/performance.d.ts

@@ -0,0 +1,2 @@
+import { Tensor } from '@tensorflow/tfjs-core';
+export default function performanceTest(fn: () => Tensor, iterations?: number, allowPromise?: boolean): Promise<number>;

package/dist/utilities/performance.js

@@ -0,0 +1,16 @@
+import { t as s } from "../index-UdZhlibC.js";
+async function f(e, o = 10, r = !1) {
+  for (let t = 0; t < 100; t++) {
+    const a = r ? await e() : s(e);
+    t === 99 && await a.data(), a.dispose();
+  }
+  const n = performance.now();
+  for (let t = 0; t < o; t++) {
+    const a = r ? await e() : s(e);
+    t === o - 1 && await a.data(), a.dispose();
+  }
+  return (performance.now() - n) / o;
+}
+export {
+  f as default
+};

package/dist/utilities/profile.d.ts

@@ -1,6 +1,7 @@
 import { MemoryInfo } from '@tensorflow/tfjs-core';
 export interface ExtendedMemoryInfo extends MemoryInfo {
     numBytesInGPUAllocated?: number;
+    numBytesAllocatedInGPU?: number;
 }
 export default class MemoryProfiler {
     private log;

package/dist/utilities/profile.js

@@ -1,5 +1,5 @@
-import { m as a } from "../index-BAzbokzv.js";
-const m = 1024 * 1024;
+import { m as a } from "../index-UdZhlibC.js";
+const s = 1024 * 1024;
 class l {
   log = /* @__PURE__ */ new Map();
   maxMemory = 0;
@@ -20,14 +20,17 @@ class l {
       console.warn("MemoryProfiler: endMemory called without matching startMemory");
       return;
     }
-    const e = a(), s = this.lastMemInfo.pop(), t = (e.numBytesInGPUAllocated ?? e.numBytes) - (s?.numBytes ?? s?.numBytesInGPUAllocated ?? 0);
-    this.log.set(o, Math.max(this.log.get(o) || 0, t)), t > this.maxMemory && (this.maxMemory = t, this.maxLabel = o), this.peakMemory = Math.max(this.peakMemory, e.numBytesInGPUAllocated ?? e.numBytes);
+    const e = a(), t = this.lastMemInfo.pop(), m = (e.numBytesInGPUAllocated ?? e.numBytesAllocatedInGPU ?? e.numBytes) - (t?.numBytesInGPUAllocated ?? t?.numBytesAllocatedInGPU ?? t?.numBytes ?? 0);
+    this.log.set(o, Math.max(this.log.get(o) || 0, m)), m > this.maxMemory && (this.maxMemory = m, this.maxLabel = o), this.peakMemory = Math.max(
+      this.peakMemory,
+      e.numBytesInGPUAllocated ?? e.numBytesAllocatedInGPU ?? e.numBytes
+    );
   }
   printSummary() {
     console.log("Memory Usage Summary:");
     for (const [o, e] of this.log.entries())
-      console.log(`- ${o}: ${(e / m).toFixed(2)} MB`);
-    this.maxLabel && console.log(`Peak Memory Usage: ${(this.maxMemory / m).toFixed(2)} MB at "${this.maxLabel}"`), console.log(`Overall Peak Memory Usage: ${(this.peakMemory / m).toFixed(2)} MB`);
+      console.log(`- ${o}: ${(e / s).toFixed(2)} MB`);
+    this.maxLabel && console.log(`Peak Memory Usage: ${(this.maxMemory / s).toFixed(2)} MB at "${this.maxLabel}"`), console.log(`Overall Peak Memory Usage: ${(this.peakMemory / s).toFixed(2)} MB`);
   }
 }
 export {

package/dist/utilities/safetensors.js

@@ -1,5 +1,5 @@
-import "../index-BAzbokzv.js";
-import { t as y } from "../tensor-Xyi595sG.js";
+import "../index-UdZhlibC.js";
+import { t as y } from "../tensor-Do9PKbIE.js";
 function l(t) {
   if (t === "float32") return "F32";
   if (t === "int32") return "I32";

package/dist/utilities/weights.js

@@ -1,5 +1,5 @@
-import "../index-BAzbokzv.js";
-import { t as p } from "../tensor-Xyi595sG.js";
+import "../index-UdZhlibC.js";
+import { t as p } from "../tensor-Do9PKbIE.js";
 function h(n) {
   const e = n.reduce((s, o) => s + o.length, 0), a = new Float32Array(e);
   let t = 0;

package/dist/{variable-wSS22xj5.js → variable-BTBkayv_.js}

@@ -1,4 +1,4 @@
-import { E as i } from "./index-BAzbokzv.js";
+import { E as i } from "./index-UdZhlibC.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.