@genai-fi/nanogpt 0.18.1 → 0.19.1

package/dist/loader/oldZipLoad.js

@@ -1,22 +1,22 @@
-import { aa as y } from "../index-DSGwv2Yx.js";
-import "../random_width-B_fVXhGx.js";
-import "../zeros_like-rOHr54NY.js";
+import { aa as y } from "../index-CUXkjxiT.js";
+import "../random_width-DN5ZtQkM.js";
+import "../zeros_like-CWjDdwr-.js";
 import "../Generator.js";
 import "../index-Cp39cXWe.js";
-import "../dataset-DlqAN81i.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-CC5RjmKU.js";
+import "../rope-DcrZM_e6.js";
 import "../ops/cpu/appendCache.js";
 import "../ops/webgl/appendCache.js";
 import "../ops/grads/softmax16.js";
-import "../matMul16-BIT70Vya.js";
+import "../matMul16-BcVC_E62.js";
 import "../ops/webgl/matMul16.js";
 import "../ops/cpu/matMul16.js";
-import "../pack16-WlOSOuZA.js";
+import "../pack16-bqltoUlR.js";
 import "../ops/transpose16.js";
 import "../ops/reshape16.js";
 import "../ops/cpu/qkv.js";
@@ -44,11 +44,11 @@ import "../ops/cpu/scatterSub.js";
 import "../ops/webgl/scatterSub.js";
 import "../papaparse.min-C0cScC2i.js";
 import "../ops/cpu/matMulGelu.js";
-import "../matMulGelu-CsZnh18H.js";
+import "../matMulGelu-JNLZqKQp.js";
 import "../ops/grads/matMulGelu.js";
 import "../ops/cpu/gelu.js";
 import "../ops/webgl/gelu.js";
-import "../gelu-DpTCC3eB.js";
+import "../gelu-B220X1Go.js";
 import "../ops/webgl/log.js";
 import "../checks/normRMS.js";
 import "../checks/normRMSGrad.js";

package/dist/loader/save.js

@@ -1,31 +1,38 @@
 import { z as y } from "../jszip.min-BZhlzntC.js";
-import b from "../tokeniser/CharTokeniser.js";
-import { save_safetensors as _ } from "../utilities/safetensors.js";
-import { VERSION as m } from "./load.js";
-async function d(i, a, o, t) {
-  const g = /* @__PURE__ */ new Map();
-  i.weightStore.saveWeights(g);
-  const n = new y();
-  if (t?.optimizer) {
-    const e = await t.optimizer.saveMoments();
-    n.file("optimizer.safetensors", e, { binary: !0 }), n.file("optimizer_config.json", JSON.stringify(t.optimizer.serializeConfig()), {
+import _ from "../tokeniser/CharTokeniser.js";
+import { save_safetensors as b } from "../utilities/safetensors.js";
+import { VERSION as h } from "./load.js";
+function m(i) {
+  if (i.length > 1e3) {
+    const n = Math.ceil(i.length / 1e3);
+    return i.filter((o, a) => a % n === 0 || a === i.length - 1);
+  }
+  return i;
+}
+async function u(i, n, o, a) {
+  const s = /* @__PURE__ */ new Map();
+  i.weightStore.saveWeights(s);
+  const e = new y();
+  if (a?.optimizer) {
+    const t = await a.optimizer.saveMoments();
+    e.file("optimizer.safetensors", t, { binary: !0 }), e.file("optimizer_config.json", JSON.stringify(a.optimizer.serializeConfig()), {
       binary: !1
     });
   }
-  t?.trainingLog && n.file("training_log.json", JSON.stringify(t.trainingLog, void 0, 4), {
+  a?.trainingLog && e.file("training_log.json", JSON.stringify(m(a.trainingLog), void 0, 4), {
     binary: !1
   });
-  const s = {};
-  g.forEach((e, f) => {
-    e.length === 1 && (s[f] = e[0]);
+  const g = {};
+  s.forEach((t, f) => {
+    t.length === 1 && (g[f] = t[0]);
   });
-  const c = await _(s);
-  n.file("model.safetensors", c, { binary: !0 });
+  const c = await b(g);
+  e.file("model.safetensors", c, { binary: !0 });
   const p = i.config.modelType;
   let r;
   p === "GenAI_NanoGPT_v1" ? r = {
     model_type: "GenAI_NanoGPT_v1",
-    vocab_size: a.getVocab().length,
+    vocab_size: n.getVocab().length,
     hidden_size: i.config.nEmbed,
     num_hidden_layers: i.config.nLayer,
     num_attention_heads: i.config.nHead,
@@ -34,7 +41,7 @@ async function d(i, a, o, t) {
     useRope: i.config.useRope
   } : r = {
     model_type: "GenAI_NanoGPT_v2",
-    vocab_size: a.getVocab().length,
+    vocab_size: n.getVocab().length,
     hidden_size: i.config.nEmbed,
     num_hidden_layers: i.config.nLayer,
     num_attention_heads: i.config.nHead,
@@ -43,11 +50,11 @@ async function d(i, a, o, t) {
     loraConfig: i.config.loraConfig ? Object.fromEntries(i.config.loraConfig) : void 0,
     loraName: i.config.loraName,
     windowSize: i.config.windowSize
-  }, n.file("config.json", JSON.stringify(r, void 0, 4), {
+  }, e.file("config.json", JSON.stringify(r, void 0, 4), {
     binary: !1
   });
   const l = {
-    version: m,
+    version: h,
     application: "@genai-fi/nanogpt",
     meta: o?.metadata,
     name: o?.name,
@@ -59,25 +66,25 @@ async function d(i, a, o, t) {
     generationSettings: i.metaData?.generationSettings || void 0,
     actionLog: i.metaData?.actionLog || void 0
   };
-  if (n.file("meta.json", JSON.stringify(l, void 0, 4), {
+  if (e.file("meta.json", JSON.stringify(l, void 0, 4), {
     binary: !1
-  }), n.file(
+  }), e.file(
     "tokeniser.json",
     JSON.stringify({
-      type: a instanceof b ? "char" : "bpe",
-      vocab: a.getVocab(),
-      merges: a.getMerges(),
-      datasetID: a.datasetID,
-      id: a.id
+      type: n instanceof _ ? "char" : "bpe",
+      vocab: n.getVocab(),
+      merges: n.getMerges(),
+      datasetID: n.datasetID,
+      id: n.id
     }),
     {
       binary: !1
     }
   ), o?.files)
-    for (const [e, f] of Object.entries(o.files))
-      n.file(e, JSON.stringify(f), { binary: !1 });
-  return n.generateAsync({ type: "blob" });
+    for (const [t, f] of Object.entries(o.files))
+      e.file(t, JSON.stringify(f), { binary: !1 });
+  return e.generateAsync({ type: "blob" });
 }
 export {
-  d as saveModel
+  u as saveModel
 };

package/dist/main.js

@@ -1,9 +1,9 @@
-import "./index-DSGwv2Yx.js";
-import "./random_width-B_fVXhGx.js";
-import "./zeros_like-rOHr54NY.js";
+import "./index-CUXkjxiT.js";
+import "./random_width-DN5ZtQkM.js";
+import "./zeros_like-CWjDdwr-.js";
 import { default as io } from "./Generator.js";
 import "./index-Cp39cXWe.js";
-import "./dataset-DlqAN81i.js";
+import "./dataset-CGGp1z9P.js";
 import { default as fo } from "./models/NanoGPTV1.js";
 import { default as lo } from "./TeachableLLM.js";
 import { default as uo } from "./tokeniser/CharTokeniser.js";
@@ -32,15 +32,15 @@ import "./ops/webgl/qkv.js";
 import "./ops/grads/qkv.js";
 import "./ops/cpu/rope.js";
 import "./ops/webgl/rope.js";
-import "./rope-CC5RjmKU.js";
+import "./rope-DcrZM_e6.js";
 import "./ops/cpu/appendCache.js";
 import "./ops/webgl/appendCache.js";
 import "./ops/cpu/matMulGelu.js";
-import "./matMulGelu-CsZnh18H.js";
+import "./matMulGelu-JNLZqKQp.js";
 import "./ops/grads/matMulGelu.js";
 import "./ops/cpu/gelu.js";
 import "./ops/webgl/gelu.js";
-import "./gelu-DpTCC3eB.js";
+import "./gelu-B220X1Go.js";
 import "./ops/cpu/normRMS.js";
 import "./ops/webgl/normRMS.js";
 import "./ops/grads/normRMS.js";
@@ -49,9 +49,9 @@ import "./ops/cpu/adamMoments.js";
 import "./ops/webgl/adamMoments.js";
 import "./ops/cpu/adamAdjust.js";
 import "./ops/webgl/adamAdjust.js";
-import { u as e, p as m } from "./pack16-WlOSOuZA.js";
+import { u as e, p as m } from "./pack16-bqltoUlR.js";
 import "./ops/grads/softmax16.js";
-import "./matMul16-BIT70Vya.js";
+import "./matMul16-BcVC_E62.js";
 import "./ops/webgl/matMul16.js";
 import "./ops/cpu/matMul16.js";
 import "./ops/transpose16.js";

package/dist/{matMul16-BIT70Vya.js → matMul16-BcVC_E62.js}

@@ -1,9 +1,9 @@
-import { c as y, e as h } from "./index-DSGwv2Yx.js";
+import { e as y, h } from "./index-CUXkjxiT.js";
 import "./ops/webgl/matMul16.js";
 import "./ops/cpu/matMul16.js";
 import { isPackedTensor as v } from "./utilities/packed.js";
-import { p as g } from "./pack16-WlOSOuZA.js";
-import { d as k } from "./gelu-DpTCC3eB.js";
+import { p as g } from "./pack16-bqltoUlR.js";
+import { d as k } from "./gelu-B220X1Go.js";
 import { transpose16 as S } from "./ops/transpose16.js";
 import { reshape16 as w } from "./ops/reshape16.js";
 import { mul16 as D } from "./ops/mul16.js";

package/dist/{matMulGelu-CsZnh18H.js → matMulGelu-JNLZqKQp.js}

@@ -1,11 +1,11 @@
-import { a as C, t as R, e as I, N as G, _ as L, v as F, am as U } from "./index-DSGwv2Yx.js";
-import { r as M } from "./Reshape-BZC-ebeR.js";
-import { u as H } from "./gpgpu_math-3bCb5ooU.js";
-import { m as B } from "./mat_mul-DP86qZtZ.js";
+import { c as C, t as R, h as I, U as G, _ as L, x as U, am as F } from "./index-CUXkjxiT.js";
+import { r as M } from "./Reshape-BW__R4mZ.js";
+import { u as H } from "./gpgpu_math-BwvV12df.js";
+import { m as B } from "./mat_mul-DhG0Newp.js";
 class W {
   constructor(e, s, a, n = !1, o = !1, r = !1, i = null, u = !1, l = !1) {
     this.variableNames = ["matrixA", "matrixB"], this.packedInputs = !0, this.packedOutput = !0, this.outputShape = a, this.enableShapeUniforms = H(this.outputShape.length);
-    const p = n ? e[1] : e[2], h = Math.ceil(p / 2), d = n ? "i * 2, rc.y" : "rc.y, i * 2", x = o ? "rc.z, i * 2" : "i * 2, rc.z", b = n ? ["a.xxyy", "a.zzww"] : ["a.xxzz", "a.yyww"], m = o ? ["b.xzxz", "b.ywyw"] : ["b.xyxy", "b.zwzw"];
+    const p = n ? e[1] : e[2], h = Math.ceil(p / 2), d = n ? "i * 2, rc.y" : "rc.y, i * 2", b = o ? "rc.z, i * 2" : "i * 2, rc.z", x = n ? ["a.xxyy", "a.zzww"] : ["a.xxzz", "a.yyww"], m = o ? ["b.xzxz", "b.ywyw"] : ["b.xyxy", "b.zwzw"];
     let c = "", g = "";
     i && (u ? c = `vec4 activation(vec4 a) {
           vec4 b = getPreluActivationWeightsAtOutCoords();
@@ -30,12 +30,12 @@ class W {
         int batchB = ${v};
         for (int i = 0; i < ${h}; i++) {
           vec4 a = getMatrixA(batchA, ${d});
-          vec4 b = getMatrixB(batchB, ${x});
+          vec4 b = getMatrixB(batchB, ${b});
 
           // These swizzled products need to be separately added.
           // See: https://github.com/tensorflow/tfjs/issues/1735
-          result += (${b[0]} * ${m[0]});
-          result += (${b[1]} * ${m[1]});
+          result += (${x[0]} * ${m[0]});
+          result += (${x[1]} * ${m[1]});
         }
         return result;
       }
@@ -81,7 +81,7 @@ const S = 0.7978845608028654, w = 0.044715, j = `
     vec4 dgelu = 0.5 * (1.0 + t) + 0.5 * a * sech2 * du_dx;
     return dgelu * b;
 `, se = 1e3;
-function N({
+function O({
   a: t,
   b: e,
   transposeA: s,
@@ -90,24 +90,24 @@ function N({
   activationSnippet: o,
   multiplier: r
 }) {
-  const i = t.shape.length, u = e.shape.length, l = s ? t.shape[i - 2] : t.shape[i - 1], p = a ? e.shape[u - 1] : e.shape[u - 2], h = s ? t.shape[i - 1] : t.shape[i - 2], d = a ? e.shape[u - 2] : e.shape[u - 1], x = t.shape.slice(0, -2), b = e.shape.slice(0, -2), m = G(x), c = G(b), $ = L(t.shape.slice(0, -2), e.shape.slice(0, -2)).concat([h, d]);
-  F(
+  const i = t.shape.length, u = e.shape.length, l = s ? t.shape[i - 2] : t.shape[i - 1], p = a ? e.shape[u - 1] : e.shape[u - 2], h = s ? t.shape[i - 1] : t.shape[i - 2], d = a ? e.shape[u - 2] : e.shape[u - 1], b = t.shape.slice(0, -2), x = e.shape.slice(0, -2), m = G(b), c = G(x), $ = L(t.shape.slice(0, -2), e.shape.slice(0, -2)).concat([h, d]);
+  U(
     l === p,
     () => `Error in matMul: inner shapes (${l}) and (${p}) of Tensors with shapes ${t.shape} and ${e.shape} and transposeA=${s} and transposeB=${a} must match.`
   );
-  const f = s ? [m, l, h] : [m, h, l], v = a ? [c, d, p] : [c, p, d], A = M({ inputs: { x: t }, backend: n, attrs: { shape: f } }), y = M({ inputs: { x: e }, backend: n, attrs: { shape: v } }), D = [A, y], O = Math.max(m, c), _ = o, E = U(t.dtype, e.dtype), T = new W(
+  const f = s ? [m, l, h] : [m, h, l], v = a ? [c, d, p] : [c, p, d], A = M({ inputs: { x: t }, backend: n, attrs: { shape: f } }), y = M({ inputs: { x: e }, backend: n, attrs: { shape: v } }), D = [A, y], _ = Math.max(m, c), E = o, N = F(t.dtype, e.dtype), T = new W(
     f,
     v,
-    [O, h, d],
+    [_, h, d],
     s,
     a,
     !1,
-    _,
+    E,
     !!r,
     !1
   ), k = [A, y];
   r && k.push(r);
-  const z = n.runWebGLProgram(T, k, E), K = M({ inputs: { x: z }, backend: n, attrs: { shape: $ } });
+  const z = n.runWebGLProgram(T, k, N), K = M({ inputs: { x: z }, backend: n, attrs: { shape: $ } });
   D.push(z);
   for (const P of D)
     n.disposeIntermediateTensorInfo(P);
@@ -117,7 +117,7 @@ function J(t) {
   const { inputs: e, backend: s } = t, { x: a, kernel: n } = e;
   if (a === void 0 || n === void 0)
     throw new Error("BatchMatMul requires two input tensors.");
-  return N({
+  return O({
     a,
     b: n,
     transposeA: !1,
@@ -136,7 +136,7 @@ function V(t) {
   const { dy: e, x: s, kernel: a } = t.inputs, n = t.backend;
   return R(() => {
     const o = I().makeTensorFromTensorInfo(
-      N({
+      O({
         a: s,
         b: a,
         transposeA: !1,
@@ -158,6 +158,6 @@ C(X);
 export {
   W as M,
   se as a,
-  N as b,
+  O as b,
   J as c
 };

package/dist/mat_mul-DhG0Newp.js

@@ -0,0 +1,11 @@
+import { o as m, q as s, B as c, E as M, D as p } from "./index-CUXkjxiT.js";
+function f(e, o, n = !1, l = !1) {
+  let a = s(e, "a", "matMul"), t = s(o, "b", "matMul");
+  [a, t] = c(a, t);
+  const r = { a, b: t }, u = { transposeA: n, transposeB: l };
+  return M.runKernel(p, r, u);
+}
+const i = /* @__PURE__ */ m({ matMul_: f });
+export {
+  i as m
+};

package/dist/mod-CSdCpRjf.js

@@ -0,0 +1,11 @@
+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.js

@@ -2,11 +2,11 @@ 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-DSGwv2Yx.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-WlOSOuZA.js";
+import { p as _, u as C } from "../pack16-bqltoUlR.js";
 const P = {
   modelType: "GenAI_NanoGPT_v1",
   vocabSize: 2e3,

package/dist/models/NanoGPTV2.js

@@ -2,10 +2,10 @@ 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-DSGwv2Yx.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-WlOSOuZA.js";
+import { p as C, u as _ } from "../pack16-bqltoUlR.js";
 const E = {
   modelType: "GenAI_NanoGPT_v2",
   vocabSize: 2e3,

package/dist/models/model.d.ts

@@ -9,6 +9,7 @@ export interface ModelForwardAttributes extends ForwardAttributes {
     attentionScores?: AttentionScores;
     seed?: number;
     skipLogits?: boolean;
+    ropePositionOffset?: number;
 }
 export interface TrainingState {
     steps: number;

package/dist/models/model.js

@@ -1,23 +1,23 @@
 import i from "../layers/BaseLayer.js";
-import "../index-DSGwv2Yx.js";
-import "../random_width-B_fVXhGx.js";
-import "../zeros_like-rOHr54NY.js";
+import "../index-CUXkjxiT.js";
+import "../random_width-DN5ZtQkM.js";
+import "../zeros_like-CWjDdwr-.js";
 import "../Generator.js";
 import "../index-Cp39cXWe.js";
-import "../dataset-DlqAN81i.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-CC5RjmKU.js";
+import "../rope-DcrZM_e6.js";
 import "../ops/cpu/appendCache.js";
 import "../ops/webgl/appendCache.js";
 import "../ops/grads/softmax16.js";
-import "../matMul16-BIT70Vya.js";
+import "../matMul16-BcVC_E62.js";
 import "../ops/webgl/matMul16.js";
 import "../ops/cpu/matMul16.js";
-import "../pack16-WlOSOuZA.js";
+import "../pack16-bqltoUlR.js";
 import "../ops/transpose16.js";
 import "../ops/reshape16.js";
 import "../ops/cpu/qkv.js";
@@ -42,11 +42,11 @@ 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-CsZnh18H.js";
+import "../matMulGelu-JNLZqKQp.js";
 import "../ops/grads/matMulGelu.js";
 import "../ops/cpu/gelu.js";
 import "../ops/webgl/gelu.js";
-import "../gelu-DpTCC3eB.js";
+import "../gelu-B220X1Go.js";
 import "../ops/webgl/log.js";
 import "../checks/normRMS.js";
 import "../checks/normRMSGrad.js";

package/dist/{not_equal-CkQKkKZy.js → not_equal-hurPF26l.js}

@@ -1,10 +1,10 @@
-import { o as c, n as e, _ as d, E as u, aL as b, aM as E, aN as _, aO as g, ah as S, a1 as w, h as K, aP as q, z as M, aQ as T } from "./index-DSGwv2Yx.js";
-import { b as h } from "./broadcast_to-C_EJTVTZ.js";
-import { e as N } from "./axis_util-QWWgLjut.js";
-import { r as f } from "./reshape-BIN71H3p.js";
-import { s as A } from "./sum-D39FeU5h.js";
+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 = d(d(r.shape, a.shape), t.shape), l = h(r, p), m = h(a, p), x = h(t, p), i = {
+  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
@@ -22,40 +22,40 @@ function L(s) {
   return u.runKernel(_, n);
 }
 const P = /* @__PURE__ */ c({ exp_: L });
-function k(s) {
+function j(s) {
   const n = { x: e(s, "x", "log", "float32") };
   return u.runKernel(g, n);
 }
-const v = /* @__PURE__ */ c({ log_: k });
-function y(s, o = null, n = !1) {
+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 = v(m), i = K(f(r, x.shape), x);
+  ), 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_: y });
-function z(s, o = null, n = !1) {
+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_: z });
+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), d(n.shape, a.shape);
+  [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,
-  v as b,
+  k as b,
   P as e,
   J as l,
   I as m,

package/dist/{ones-DbVB5N58.js → ones-BytntneX.js}

@@ -1,6 +1,6 @@
-import { q as n, Q as t, N as m, E as i } from "./index-DSGwv2Yx.js";
-import { c as f } from "./complex-3DpPEG9B.js";
-import { z as c } from "./zeros-Bw0puq_w.js";
+import { u as n, V as t, U as m, E as i } from "./index-CUXkjxiT.js";
+import { c as f } from "./complex-DI35Q-gW.js";
+import { z as c } from "./zeros-DvZpK8s6.js";
 function l(o, r = "float32") {
   if (n(o), r === "complex64") {
     const s = l(o, "float32"), a = c(o, "float32");

package/dist/ops/adamAdjust.js

@@ -1,8 +1,8 @@
-import { e as u } from "../index-DSGwv2Yx.js";
+import { h as u } from "../index-CUXkjxiT.js";
 import "./cpu/adamAdjust.js";
 import "./webgl/adamAdjust.js";
-function s(r, t, e, n, m, o, i = 0) {
-  return u().runKernel("AdamAdjust", { moments: r, value: t }, { beta1: e, beta2: n, epsilon: m, learningRate: o, weightDecay: i });
+function s(r, t, n, e, m, o, i = 0) {
+  return u().runKernel("AdamAdjust", { moments: r, value: t }, { beta1: n, beta2: e, epsilon: m, learningRate: o, weightDecay: i });
 }
 export {
   s as adamAdjust

package/dist/ops/adamMoments.js

@@ -1,8 +1,8 @@
-import { e as t } from "../index-DSGwv2Yx.js";
+import { h as t } from "../index-CUXkjxiT.js";
 import "./cpu/adamMoments.js";
 import "./webgl/adamMoments.js";
-function s(e, n, r, m, o) {
-  return t().runKernel("AdamMoments", { moments: e, gradient: n, scaling: o }, { beta1: r, beta2: m });
+function s(n, r, e, m, o) {
+  return t().runKernel("AdamMoments", { moments: n, gradient: r, scaling: o }, { beta1: e, beta2: m });
 }
 export {
   s as adamMoments

package/dist/ops/add16.js

@@ -1,4 +1,4 @@
-import { h as t, e as o } from "../index-DSGwv2Yx.js";
+import { j as t, h as o } from "../index-CUXkjxiT.js";
 import { isPackedTensor as n } from "../utilities/packed.js";
 import "./grads/add16.js";
 function m(r, e) {

package/dist/ops/appendCache.js

@@ -1,21 +1,21 @@
-import { e as a } from "../index-DSGwv2Yx.js";
+import { h as e } from "../index-CUXkjxiT.js";
 import "./cpu/appendCache.js";
 import "./webgl/appendCache.js";
 import { isPackedTensor as c } from "../utilities/packed.js";
-import { c as t } from "../concat-Dqk7Xk7h.js";
-import { z as f } from "../zeros-Bw0puq_w.js";
+import { c as t } from "../concat-CAQpCret.js";
+import { z as f } from "../zeros-DvZpK8s6.js";
 function C(r, o, n, p) {
   if (!p) {
-    const e = r.shape[2], s = c(r);
+    const s = r.shape[2], a = c(r);
     return t(
       [
         r,
-        f([r.shape[0], r.shape[1], o - e, r.shape[3]], s ? "int32" : r.dtype)
+        f([r.shape[0], r.shape[1], o - s, r.shape[3]], a ? "int32" : r.dtype)
       ],
       2
     );
   }
-  return a().runKernel("AppendCache", { cache: p, item: r }, { maxSize: o, pastLen: n });
+  return e().runKernel("AppendCache", { cache: p, item: r }, { maxSize: o, pastLen: n });
 }
 export {
   C as appendCache

package/dist/ops/attentionMask.js

@@ -1,9 +1,9 @@
-import { e as r } from "../index-DSGwv2Yx.js";
+import { h as e } from "../index-CUXkjxiT.js";
 import "./cpu/attentionMask.js";
 import "./webgl/attentionMask.js";
 import "./grads/attentionMask.js";
-function u(t, n, e, o) {
-  return r().runKernel("AttentionMask", { q: t, k: n }, { divisor: e, pastLen: o || 0 });
+function u(t, n, o, r) {
+  return e().runKernel("AttentionMask", { q: t, k: n }, { divisor: o, pastLen: r || 0 });
 }
 export {
   u as attentionMask

package/dist/ops/concat16.js

@@ -1,8 +1,8 @@
 import { isPackedTensor as o } from "../utilities/packed.js";
-import { e } from "../index-DSGwv2Yx.js";
-import { c } from "../concat-Dqk7Xk7h.js";
+import { h as c } from "../index-CUXkjxiT.js";
+import { c as e } from "../concat-CAQpCret.js";
 function p(r, n) {
-  return o(r[0]) ? e().runKernel("Concat16", r, { axis: n ?? -1 }) : c(r, n);
+  return o(r[0]) ? c().runKernel("Concat16", r, { axis: n ?? -1 }) : e(r, n);
 }
 export {
   p as concat16

package/dist/ops/cpu/adamAdjust.js

@@ -1,18 +1,18 @@
-import { a as h, d as t, h as i, m as k, i as z } from "../../index-DSGwv2Yx.js";
-function A(a) {
-  const { moments: s, value: r } = a.inputs, { beta1: l, beta2: m, epsilon: u, learningRate: d } = a.attrs, e = s.shape.length, c = new Array(e).fill(0), n = s.shape.slice();
+import { c as k, i as t, j as i, m as z, l as A } from "../../index-CUXkjxiT.js";
+function C(c) {
+  const { moments: s, value: r } = c.inputs, { beta1: l, beta2: m, epsilon: u, learningRate: d } = c.attrs, e = s.shape.length, a = new Array(e).fill(0), n = s.shape.slice();
   n[e - 1] = 1;
-  const o = c.slice();
+  const o = a.slice();
   o[e - 1] = 1;
-  const b = n.slice(), p = s.slice(c, n).squeeze([e - 1]), M = s.slice(o, b).squeeze([e - 1]), g = t(p, l), f = t(M, m);
+  const b = n.slice(), p = s.slice(a, n).squeeze([e - 1]), M = s.slice(o, b).squeeze([e - 1]), g = t(p, l), f = t(M, m);
   return i(
-    k(t(g, i(z(f), u ?? 1e-8)), -d),
+    z(t(g, i(A(f), u ?? 1e-8)), -d),
     r
   );
 }
-const C = {
+const h = {
   kernelName: "AdamAdjust",
   backendName: "cpu",
-  kernelFunc: A
+  kernelFunc: C
 };
-h(C);
+k(h);

package/dist/ops/cpu/adamMoments.js

@@ -1,11 +1,11 @@
-import { a as p } from "../../index-DSGwv2Yx.js";
-import { s as b } from "../../stack-z6QE7kmP.js";
+import { c as p } from "../../index-CUXkjxiT.js";
+import { s as b } from "../../stack-DudVrtmG.js";
 function f(t) {
-  const { moments: n, gradient: c } = t.inputs, { beta1: o, beta2: a } = t.attrs, e = n.shape.length, m = new Array(e).fill(0), s = n.shape.slice();
+  const { moments: n, gradient: c } = t.inputs, { beta1: o, beta2: m } = t.attrs, e = n.shape.length, a = new Array(e).fill(0), s = n.shape.slice();
   s[e - 1] = 1;
-  const i = m.slice();
+  const i = a.slice();
   i[e - 1] = 1;
-  const r = s.slice(), l = n.slice(m, s).squeeze([e - 1]), u = n.slice(i, r).squeeze([e - 1]), M = l.mul(o).add(c.mul(1 - o)), d = u.mul(a).add(c.square().mul(1 - a));
+  const r = s.slice(), l = n.slice(a, s).squeeze([e - 1]), u = n.slice(i, r).squeeze([e - 1]), M = l.mul(o).add(c.mul(1 - o)), d = u.mul(m).add(c.square().mul(1 - m));
   return b([M, d], -1);
 }
 const g = {