@genai-fi/nanogpt 0.3.1 → 0.4.0

package/dist/axis_util-DeydwOoC.js

@@ -0,0 +1,69 @@
+import { j as c } from "./index-XjBAhiFO.js";
+/**
+ * @license
+ * Copyright 2017 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 i(e, n) {
+  for (let t = 0; t < e.length; ++t)
+    if (e[e.length - t - 1] !== n - 1 - t)
+      return !1;
+  return !0;
+}
+function p(e, n, t) {
+  const r = e.length + n.length, s = [];
+  let o = 0, f = 0;
+  for (let u = 0; u < r; u++)
+    t.indexOf(u) === -1 ? s.push(e[o++]) : s.push(n[f++]);
+  return s;
+}
+function a(e, n) {
+  const t = [], r = e.length;
+  for (let o = 0; o < r; o++)
+    n.indexOf(o) === -1 && t.push(e[o]);
+  const s = n.map((o) => e[o]);
+  return [t, s];
+}
+function m(e, n) {
+  const t = n.map((r) => 1);
+  return p(e, t, n);
+}
+function d(e, n, t) {
+  c(i(n, t), () => `${e} supports only inner-most axes for now. Got axes ${n} and rank-${t} input.`);
+}
+function h(e, n) {
+  if (i(e, n))
+    return null;
+  const t = [];
+  for (let r = 0; r < n; ++r)
+    e.indexOf(r) === -1 && t.push(r);
+  return e.forEach((r) => t.push(r)), t;
+}
+function g(e) {
+  return e.map((n, t) => [t, n]).sort((n, t) => n[1] - t[1]).map((n) => n[0]);
+}
+function x(e, n) {
+  const t = [];
+  for (let r = n - e; r < n; ++r)
+    t.push(r);
+  return t;
+}
+export {
+  x as a,
+  d as b,
+  a as c,
+  g as d,
+  m as e,
+  h as g
+};

package/dist/{concat-BIZS_td9.js → concat-DS_qH7MI.js}

@@ -1,4 +1,4 @@
-import { o as s, j as a, i, u as p, E as l, C as u } from "./index-pWA4_lUh.js";
+import { o as s, j as a, i, w as p, E as l, C as f } from "./index-XjBAhiFO.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.
@@ -15,7 +15,7 @@ import { o as s, j as a, i, u as p, E as l, C as u } from "./index-pWA4_lUh.js";
  * limitations under the License.
  * =============================================================================
  */
-function f(o, e = 0) {
+function h(o, e = 0) {
   a(o.length >= 1, () => "Pass at least one tensor to concat");
   const t = i(o, "tensors", "concat", "string_or_numeric");
   if (t[0].dtype === "complex64" && t.forEach((n) => {
@@ -25,9 +25,9 @@ function f(o, e = 0) {
   }), t.length === 1)
     return p(t[0]);
   const r = t, c = { axis: e };
-  return l.runKernel(u, r, c);
+  return l.runKernel(f, r, c);
 }
-const m = /* @__PURE__ */ s({ concat_: f });
+const u = /* @__PURE__ */ s({ concat_: h });
 export {
-  m as c
+  u as c
 };

package/dist/config.js

@@ -1,20 +1,19 @@
 const e = {
-  vocabSize: 50304,
-  // GPT-2 vocab size
-  blockSize: 1024,
+  vocabSize: 2e3,
+  blockSize: 128,
   // Maximum sequence length
-  nLayer: 12,
+  nLayer: 6,
   // Number of transformer layers
-  nHead: 12,
+  nHead: 4,
   // Number of attention heads
-  nEmbed: 768,
+  nEmbed: 256,
   // Embedding dimension
-  dropout: 0,
+  dropout: 0.1,
   // Dropout probability
   biasInLinear: !1,
   biasInLayerNorm: !1,
   mlpFactor: 4,
-  useRope: !1
+  useRope: !0
   // Use Rotary Position Embeddings
 };
 export {

package/dist/{gather-BPGW8RsB.js → gather-BUmJIS8n.js}

@@ -1,4 +1,4 @@
-import { o as h, h as t, E as g, G as p } from "./index-pWA4_lUh.js";
+import { o as h, h as t, E as g, G as p } from "./index-XjBAhiFO.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.