@genai-fi/nanogpt 0.2.9 → 0.2.11

package/dist/training/sparseCrossEntropy.js

@@ -1,7 +1,8 @@
-import { gatherSub as w } from "../ops/gatherSub.js";
-import { scatterSub as K } from "../ops/scatterSub.js";
-import { o as l, c as d, E as f, M as _, t as z, L as I, u as N, a as E, v as M, w as T, e as m, x as g, y as $, z as S } from "../index-Dsg28SG6.js";
-import { s as F, r as b } from "../sum-NWazHI7f.js";
+import { gatherSub as _ } from "../ops/gatherSub.js";
+import { scatterSub as v } from "../ops/scatterSub.js";
+import { o as l, d, E as f, M as w, y as I, L as z, z as N, b as E, D as F, F as M, e as m, H as g, I as $, J as S } from "../index-YPKosni4.js";
+import { r as b } from "../reshape-DmnmKT6r.js";
+import { s as T } from "../sum-D7fu15XL.js";
 /**
  * @license
  * Copyright 2017 Google LLC. All Rights Reserved.
@@ -18,16 +19,16 @@ import { s as F, r as b } from "../sum-NWazHI7f.js";
  * limitations under the License.
  * =============================================================================
  */
-function P(n, s, t) {
+function D(n, s, t) {
   const a = n.length + s.length, e = [];
   let r = 0, c = 0;
   for (let o = 0; o < a; o++)
     t.indexOf(o) === -1 ? e.push(n[r++]) : e.push(s[c++]);
   return e;
 }
-function A(n, s) {
+function P(n, s) {
   const t = s.map((a) => 1);
-  return P(n, t, s);
+  return D(n, t, s);
 }
 /**
  * @license
@@ -45,11 +46,11 @@ function A(n, s) {
  * limitations under the License.
  * =============================================================================
  */
-function D(n, s = null, t = !1) {
+function A(n, s = null, t = !1) {
   const e = { x: d(n, "x", "max") }, r = { reductionIndices: s, keepDims: t };
-  return f.runKernel(_, e, r);
+  return f.runKernel(w, e, r);
 }
-const L = /* @__PURE__ */ l({ max_: D });
+const L = /* @__PURE__ */ l({ max_: A });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -66,11 +67,11 @@ const L = /* @__PURE__ */ l({ max_: D });
  * limitations under the License.
  * =============================================================================
  */
-function O(n) {
+function H(n) {
   const t = { x: d(n, "x", "exp") };
-  return f.runKernel(z, t);
+  return f.runKernel(I, t);
 }
-const W = /* @__PURE__ */ l({ exp_: O });
+const J = /* @__PURE__ */ l({ exp_: H });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -87,11 +88,11 @@ const W = /* @__PURE__ */ l({ exp_: O });
  * limitations under the License.
  * =============================================================================
  */
-function j(n) {
+function O(n) {
   const t = { x: d(n, "x", "log", "float32") };
-  return f.runKernel(I, t);
+  return f.runKernel(z, t);
 }
-const q = /* @__PURE__ */ l({ log_: j });
+const W = /* @__PURE__ */ l({ log_: O });
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -108,20 +109,20 @@ const q = /* @__PURE__ */ l({ log_: j });
  * limitations under the License.
  * =============================================================================
  */
-function B(n, s = null, t = !1) {
+function j(n, s = null, t = !1) {
   const a = d(n, "x", "logSumExp"), e = N(s, a.shape), r = L(
     a,
     e,
     !0
     /* keepDims */
-  ), c = E(a, r), o = W(c), p = F(o, e), u = q(p), i = M(b(r, u.shape), u);
+  ), c = E(a, r), o = J(c), p = T(o, e), u = W(p), x = F(b(r, u.shape), u);
   if (t) {
-    const h = A(i.shape, e);
-    return b(i, h);
+    const h = P(x.shape, e);
+    return b(x, h);
   }
-  return i;
+  return x;
 }
-const H = /* @__PURE__ */ l({ logSumExp_: B });
+const q = /* @__PURE__ */ l({ logSumExp_: j });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -138,21 +139,21 @@ const H = /* @__PURE__ */ l({ logSumExp_: B });
  * limitations under the License.
  * =============================================================================
  */
-function J(n, s = -1) {
+function B(n, s = -1) {
   const t = d(n, "logits", "softmax", "float32");
   if (s === -1 && (s = t.rank - 1), s !== t.rank - 1)
     throw Error(`Softmax along a non-last dimension is not yet supported. Logits was rank ${t.rank} and dim was ${s}`);
   const a = { logits: t }, e = { dim: s };
-  return f.runKernel(T, a, e);
+  return f.runKernel(M, a, e);
 }
-const Q = /* @__PURE__ */ l({ softmax_: J });
+const Q = /* @__PURE__ */ l({ softmax_: B });
 function R(n, s) {
   return $(() => {
-    const t = n.shape[n.shape.length - 1], e = n.shape.slice(0, -1).reduce((h, x) => h * x, 1), r = n.shape.length > 2 ? n.reshape([e, t]) : n, c = s.shape.length > 1 ? s.reshape([e]).cast("int32") : s.cast("int32"), o = L(r, -1, !0), p = E(r, o), u = H(p, -1);
-    return w(u, c, p);
+    const t = n.shape[n.shape.length - 1], e = n.shape.slice(0, -1).reduce((h, i) => h * i, 1), r = n.shape.length > 2 ? n.reshape([e, t]) : n, c = s.shape.length > 1 ? s.reshape([e]).cast("int32") : s.cast("int32"), o = L(r, -1, !0), p = E(r, o), u = q(p, -1);
+    return _(u, c, p);
   });
 }
-function ss() {
+function ts() {
   return m().backendName === "tensorflow" ? g((s, t, a) => {
     const e = s.shape.length > 2 ? s.reshape([-1, s.shape[s.shape.length - 1]]) : s, r = t.shape.length > 1 ? t.reshape([-1]).cast("int32") : t.cast("int32"), [c, o] = m().runKernel(
       "NativeSparseSoftmaxCrossEntropy",
@@ -163,15 +164,15 @@ function ss() {
   }) : g(
     // @ts-expect-error Invalid params
     (s, t, a) => {
-      const e = s.shape[s.shape.length - 1], c = s.shape.slice(0, -1).reduce((h, x) => h * x, 1), o = s.reshape([c, e]), p = t.reshape([c]).cast("int32"), u = R(o, p);
-      return a([o, p]), o.dispose(), p.dispose(), { value: u, gradFunc: (h, x) => $(() => {
-        const y = x[0], k = x[1], C = Q(y), G = K(C, k, h), v = S(t);
-        return [G.reshape(s.shape), v];
+      const e = s.shape[s.shape.length - 1], c = s.shape.slice(0, -1).reduce((h, i) => h * i, 1), o = s.reshape([c, e]), p = t.reshape([c]).cast("int32"), u = R(o, p);
+      return a([o, p]), o.dispose(), p.dispose(), { value: u, gradFunc: (h, i) => $(() => {
+        const y = i[0], k = i[1], C = Q(y), G = v(C, k, h), K = S(t);
+        return [G.reshape(s.shape), K];
       }) };
     }
   );
 }
 export {
-  ss as createSoftmaxCrossEntropyWithGrad,
+  ts as createSoftmaxCrossEntropyWithGrad,
   R as sparseSoftmaxCrossEntropy
 };

package/dist/utilities/profile.d.ts

@@ -0,0 +1,10 @@
+export default class MemoryProfiler {
+    private log;
+    private maxMemory;
+    private maxLabel?;
+    private lastMemInfo;
+    private peakMemory;
+    startMemory(): void;
+    endMemory(label: string): void;
+    printSummary(): void;
+}

package/dist/utilities/profile.js

@@ -0,0 +1,29 @@
+import { m as s } from "../index-YPKosni4.js";
+const m = 1024 * 1024;
+class i {
+  log = /* @__PURE__ */ new Map();
+  maxMemory = 0;
+  maxLabel;
+  lastMemInfo = [];
+  peakMemory = 0;
+  startMemory() {
+    this.lastMemInfo.push(s());
+  }
+  endMemory(e) {
+    if (this.lastMemInfo.length === 0) {
+      console.warn("MemoryProfiler: endMemory called without matching startMemory");
+      return;
+    }
+    const o = s(), t = o.numBytes - (this.lastMemInfo.pop()?.numBytes || 0);
+    this.log.set(e, Math.max(this.log.get(e) || 0, t)), t > this.maxMemory && (this.maxMemory = t, this.maxLabel = e), this.peakMemory = Math.max(this.peakMemory, o.numBytes);
+  }
+  printSummary() {
+    console.log("Memory Usage Summary:");
+    for (const [e, o] of this.log.entries())
+      console.log(`- ${e}: ${(o / 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`);
+  }
+}
+export {
+  i as default
+};

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@genai-fi/nanogpt",
-    "version": "0.2.9",
+    "version": "0.2.11",
     "type": "module",
     "main": "dist/main.js",
     "types": "dist/main.d.ts",

package/dist/sum-NWazHI7f.js

@@ -1,49 +0,0 @@
-import { o, c as a, E as u, i, j as p, S as x } from "./index-Dsg28SG6.js";
-/**
- * @license
- * Copyright 2020 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function l(n, t) {
-  const s = { x: a(n, "x", "reshape", "string_or_numeric") }, r = { shape: t };
-  return u.runKernel(i, s, r);
-}
-const h = /* @__PURE__ */ o({ reshape_: l });
-/**
- * @license
- * Copyright 2018 Google LLC. All Rights Reserved.
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- * =============================================================================
- */
-function m(n, t = null, e = !1) {
-  let s = a(n, "x", "sum");
-  s.dtype === "bool" && (s = p(s, "int32"));
-  const r = { x: s }, c = { axis: t, keepDims: e };
-  return u.runKernel(x, r, c);
-}
-const _ = /* @__PURE__ */ o({ sum_: m });
-export {
-  h as r,
-  _ as s
-};