@genai-fi/nanogpt 0.2.6 → 0.2.7

package/dist/training/sparseCrossEntropy.js

@@ -1,6 +1,6 @@
-import { gatherSub as K } from "../ops/gatherSub.js";
-import { scatterSub as _ } from "../ops/scatterSub.js";
-import { o as l, c as d, E as f, M as G, p as z, L as I, q as N, a as E, t as M, u as T, e as m, v as S, w as $, z as g } from "../index-D1SlunD-.js";
+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 _, p as z, L as I, q as N, a as E, t as M, u as T, e as m, v as g, w as $, z as S } from "../index-D1SlunD-.js";
 import { s as F, r as b } from "../sum-02UQ5Eaq.js";
 /**
  * @license
@@ -47,9 +47,9 @@ function q(n, s) {
  */
 function A(n, s = null, t = !1) {
   const e = { x: d(n, "x", "max") }, r = { reductionIndices: s, keepDims: t };
-  return f.runKernel(G, e, r);
+  return f.runKernel(_, e, r);
 }
-const k = /* @__PURE__ */ l({ max_: A });
+const L = /* @__PURE__ */ l({ max_: A });
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.
@@ -109,7 +109,7 @@ const j = /* @__PURE__ */ l({ log_: W });
  * =============================================================================
  */
 function B(n, s = null, t = !1) {
-  const a = d(n, "x", "logSumExp"), e = N(s, a.shape), r = k(
+  const a = d(n, "x", "logSumExp"), e = N(s, a.shape), r = L(
     a,
     e,
     !0
@@ -148,30 +148,30 @@ function J(n, s = -1) {
 const Q = /* @__PURE__ */ l({ softmax_: J });
 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 = k(r, -1, !0), p = E(r, o), u = H(p, -1);
-    return K(u, c, p);
+    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);
   });
 }
-function Z() {
-  return m().backendName === "tensorflow" ? S((s, t, a) => {
+function ss() {
+  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",
       { logits: e, labels: r },
       {}
     );
-    return a([o.reshape(s.shape)]), { value: c, gradFunc: (p, u) => [u[0], g(t)] };
-  }) : S(
+    return a([o.reshape(s.shape)]), { value: c, gradFunc: (p, u) => [u[0], S(t)] };
+  }) : 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], C = x[1], L = Q(y), v = _(L, C, h), w = g(t);
-        return [v, w];
+        const k = x[0], y = x[1], C = Q(k), G = K(C, y, h), v = S(t);
+        return [G.reshape(s.shape), v];
       }) };
     }
   );
 }
 export {
-  Z as createSoftmaxCrossEntropyWithGrad,
+  ss as createSoftmaxCrossEntropyWithGrad,
   R as sparseSoftmaxCrossEntropy
 };

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@genai-fi/nanogpt",
-    "version": "0.2.6",
+    "version": "0.2.7",
     "type": "module",
     "main": "dist/main.js",
     "types": "dist/main.d.ts",