@genai-fi/nanogpt 0.7.1 → 0.7.2

package/dist/ops/cpu/gatherSub.js

@@ -1,6 +1,6 @@
-import { B as u, C as c, E as g, a4 as h, f as m, c as p } from "../../index-UdZhlibC.js";
-import { r as f } from "../../range-D9CZhVlR.js";
-import { s as l } from "../../stack-ikk2Y8_P.js";
+import { B as u, C as c, E as g, a5 as h, f as m, c as p } from "../../index-BoWRt-10.js";
+import { r as f } from "../../range-CRuAh-gd.js";
+import { s as l } from "../../stack-B17UN2nn.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.

package/dist/ops/cpu/gelu.js

@@ -1,4 +1,4 @@
-import { f as t, t as d } from "../../index-UdZhlibC.js";
+import { f as t, t as d } from "../../index-BoWRt-10.js";
 const o = 0.7978845608028654, c = 0.044715;
 function m(r) {
   const { inputs: u } = r, { x: n } = u, e = n;

package/dist/ops/cpu/matMulGelu.js

@@ -1,5 +1,5 @@
-import { f as e, t as m } from "../../index-UdZhlibC.js";
-import { g as d, d as M } from "../../gelu-BFwVnd1r.js";
+import { f as e, t as m } from "../../index-BoWRt-10.js";
+import { g as d, d as M } from "../../gelu-C-dPj6Ku.js";
 function c(t) {
   const { inputs: u } = t, { x: n, kernel: l } = u, a = n, r = l;
   return m(() => {

package/dist/ops/cpu/matMulMul.js

@@ -1,4 +1,4 @@
-import { f as e, t as i } from "../../index-UdZhlibC.js";
+import { f as e, t as i } from "../../index-BoWRt-10.js";
 function n(t) {
   const { inputs: r, attrs: l } = t, { transposeA: o, transposeB: s } = l, { x: u, kernel: a, y: c } = r, m = u, k = a, M = c;
   return i(() => m.matMul(k, o, s).mul(M));

package/dist/ops/cpu/mulDropout.js

@@ -1,4 +1,4 @@
-import { f as e, a as t } from "../../index-UdZhlibC.js";
+import { f as e, a as t } from "../../index-BoWRt-10.js";
 function n(o) {
   const { inputs: r } = o, { a: l, b: u } = r;
   return console.warn("Using fallback mulDrop implementation without dropout."), t(l, u);

package/dist/ops/cpu/normRMS.js

@@ -1,4 +1,4 @@
-import { f as o, t as d } from "../../index-UdZhlibC.js";
+import { f as o, t as d } from "../../index-BoWRt-10.js";
 function i(t) {
   const { inputs: e } = t, { x: n, gamma: s } = e, r = n, a = s;
   return d(() => {

package/dist/ops/cpu/qkv.js

@@ -1,6 +1,6 @@
-import { f as q } from "../../index-UdZhlibC.js";
-import { r as o } from "../../reshape-BR0eoLYN.js";
-import { s as x } from "../../split-CRU0PjVV.js";
+import { f as q } from "../../index-BoWRt-10.js";
+import { r as o } from "../../reshape-CdBq1WJ6.js";
+import { s as x } from "../../split-B-dikLRw.js";
 function v(p) {
   const { x: c, kernel: K } = p.inputs, { heads: n } = p.attrs, [s, e, t] = c.shape, a = o(c, [s * e, t]), i = a.dot(K);
   a.dispose();

package/dist/ops/cpu/rope.js

@@ -1,8 +1,8 @@
-import { f as S } from "../../index-UdZhlibC.js";
-import { r as F } from "../../range-D9CZhVlR.js";
-import { g as I } from "../../gather-Bbh8DHhM.js";
-import { s as E } from "../../stack-ikk2Y8_P.js";
-import { c as T } from "../../concat-CbXTetof.js";
+import { f as S } from "../../index-BoWRt-10.js";
+import { r as F } from "../../range-CRuAh-gd.js";
+import { g as I } from "../../gather-CMMy2KEG.js";
+import { s as E } from "../../stack-B17UN2nn.js";
+import { c as T } from "../../concat-CsxrgovM.js";
 function U(t, c, p, o, r) {
   const n = o.shape[3], s = p;
   if (s > n) return o;

package/dist/ops/cpu/scatterSub.js

@@ -1,8 +1,8 @@
-import { B as f, F as g, C as r, E as l, a3 as N, f as b, c as S, a as h } from "../../index-UdZhlibC.js";
-import { v as D } from "../../scatter_nd_util-OjyAxku2.js";
-import { r as k } from "../../range-D9CZhVlR.js";
-import { s as v } from "../../stack-ikk2Y8_P.js";
-import { o as E } from "../../ones-C8Mfln6-.js";
+import { B as f, F as g, C as r, E as l, a4 as N, f as b, c as S, a as h } from "../../index-BoWRt-10.js";
+import { v as D } from "../../scatter_nd_util-DUstGbU1.js";
+import { r as k } from "../../range-CRuAh-gd.js";
+import { s as v } from "../../stack-B17UN2nn.js";
+import { o as E } from "../../ones-Dj0SDhHf.js";
 /**
  * @license
  * Copyright 2018 Google LLC. All Rights Reserved.

package/dist/ops/fusedSoftmax.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-UdZhlibC.js";
+import { e as t } from "../index-BoWRt-10.js";
 import "./cpu/fusedSoftmax.js";
 import "./webgl/fusedSoftmax.js";
 import "./grads/fusedSoftmax.js";

package/dist/ops/gatherSub.js

@@ -1,4 +1,4 @@
-import { e as n } from "../index-UdZhlibC.js";
+import { e as n } from "../index-BoWRt-10.js";
 import "./cpu/gatherSub.js";
 import "./webgl/gatherSub.js";
 function f(r, e, t) {

package/dist/ops/gelu.js

@@ -1,7 +1,7 @@
-import "../index-UdZhlibC.js";
+import "../index-BoWRt-10.js";
 import "./cpu/gelu.js";
 import "./webgl/gelu.js";
-import { d as e, g as i } from "../gelu-BFwVnd1r.js";
+import { d as e, g as i } from "../gelu-C-dPj6Ku.js";
 export {
   e as dGelu,
   i as gelu

package/dist/ops/grads/attentionMask.js

@@ -1,4 +1,4 @@
-import { i as m, b as i } from "../../index-UdZhlibC.js";
+import { i as m, b as i } from "../../index-BoWRt-10.js";
 import { matMulMul as u } from "../matMulMul.js";
 const p = {
   kernelName: "AttentionMask",

package/dist/ops/grads/fusedSoftmax.js

@@ -1,6 +1,6 @@
-import { i as f, a as i, c as l } from "../../index-UdZhlibC.js";
+import { i as f, a as i, c as l } from "../../index-BoWRt-10.js";
 import { mulDrop as g } from "../mulDrop.js";
-import { s as T } from "../../sum-NLYbiDag.js";
+import { s as T } from "../../sum-66ew2byf.js";
 const Y = {
   kernelName: "FusedSoftmax",
   outputsToSave: [!0],

package/dist/ops/grads/gelu.js

@@ -1,5 +1,5 @@
-import "../../index-UdZhlibC.js";
-import { a as e } from "../../gelu-BFwVnd1r.js";
+import "../../index-BoWRt-10.js";
+import { a as e } from "../../gelu-C-dPj6Ku.js";
 export {
   e as geluGradConfig
 };

package/dist/ops/grads/matMulGelu.js

@@ -1,4 +1,4 @@
-import { i as a, e as o } from "../../index-UdZhlibC.js";
+import { i as a, e as o } from "../../index-BoWRt-10.js";
 function i(e, n, r) {
   return o().runKernel("MatMulGeluGrad", { dy: e, x: n, kernel: r });
 }

package/dist/ops/grads/normRMS.js

@@ -1,4 +1,4 @@
-import { i as t, e as g } from "../../index-UdZhlibC.js";
+import { i as t, e as g } from "../../index-BoWRt-10.js";
 function i(r, a, n) {
   return g().runKernel("RMSNormGrad", { dy: r, x: a, gamma: n });
 }

package/dist/ops/grads/qkv.js

@@ -1,4 +1,4 @@
-import { i as Q } from "../../index-UdZhlibC.js";
+import { i as Q } from "../../index-BoWRt-10.js";
 const V = {
   kernelName: "QKV",
   inputsToSave: ["x", "kernel"],

package/dist/ops/grads/rope.js

@@ -1,4 +1,4 @@
-import { i, e as a } from "../../index-UdZhlibC.js";
+import { i, e as a } from "../../index-BoWRt-10.js";
 function p(n, e, s, o) {
   return a().runKernel("Rope", { x: n, sin: e, cos: s }, { pastLen: o });
 }

package/dist/ops/matMulGelu.js

@@ -1,4 +1,4 @@
-import { e as u } from "../index-UdZhlibC.js";
+import { e as u } from "../index-BoWRt-10.js";
 import "./cpu/matMulGelu.js";
 import "./webgl/matMulGelu.js";
 import "./grads/matMulGelu.js";

package/dist/ops/matMulMul.js

@@ -1,4 +1,4 @@
-import { e as u } from "../index-UdZhlibC.js";
+import { e as u } from "../index-BoWRt-10.js";
 import "./cpu/matMulMul.js";
 import "./webgl/matMulMul.js";
 function m(e, r, t, l = !1, n = !1) {

package/dist/ops/mulDrop.js

@@ -1,4 +1,4 @@
-import { e as t } from "../index-UdZhlibC.js";
+import { e as t } from "../index-BoWRt-10.js";
 import "./cpu/mulDropout.js";
 import "./webgl/mulDropout.js";
 function m(r, o, e, n) {

package/dist/ops/normRMS.js

@@ -1,4 +1,4 @@
-import { e as n } from "../index-UdZhlibC.js";
+import { e as n } from "../index-BoWRt-10.js";
 import "./cpu/normRMS.js";
 import "./webgl/normRMS.js";
 import "./grads/normRMS.js";

package/dist/ops/qkv.js

@@ -1,4 +1,4 @@
-import { e as o } from "../index-UdZhlibC.js";
+import { e as o } from "../index-BoWRt-10.js";
 import "./cpu/qkv.js";
 import "./webgl/qkv.js";
 import "./grads/qkv.js";

package/dist/ops/rope.js

@@ -1,8 +1,8 @@
-import { e as p } from "../index-UdZhlibC.js";
-import "../random_width-DbSpgl4o.js";
-import "../register_all_kernels-DnbAyBXt.js";
+import { e as p } from "../index-BoWRt-10.js";
+import "../random_width-sZORGo5k.js";
+import "../register_all_kernels-BwDSRN-f.js";
 import "../index-Tf7vU29b.js";
-import "../dataset-U3PrjwgU.js";
+import "../dataset-CtdBYwjo.js";
 import "./cpu/rope.js";
 import "./webgl/rope.js";
 import "./grads/rope.js";

package/dist/ops/scatterSub.js

@@ -1,4 +1,4 @@
-import { e as i } from "../index-UdZhlibC.js";
+import { e as i } from "../index-BoWRt-10.js";
 import "./cpu/scatterSub.js";
 import "./webgl/scatterSub.js";
 function c(t, r, e) {

package/dist/ops/webgl/adamAdjust.js

@@ -1,5 +1,5 @@
-import { r as n } from "../../Reshape-CEsEp0AI.js";
-import { f } from "../../index-UdZhlibC.js";
+import { r as n } from "../../Reshape-DH5srBP0.js";
+import { f } from "../../index-BoWRt-10.js";
 class v {
   variableNames = ["moments", "value"];
   outputShape;

package/dist/ops/webgl/adamMoments.js

@@ -1,4 +1,4 @@
-import { f as m } from "../../index-UdZhlibC.js";
+import { f as m } from "../../index-BoWRt-10.js";
 class i {
   variableNames = ["moments", "gradient"];
   outputShape;
@@ -9,12 +9,14 @@ class i {
   ];
   constructor(e) {
     this.outputShape = e;
-    const t = e.length, n = t === 1 ? "int" : `ivec${Math.min(t, 4)}`, a = t === 1 ? "coords" : `coords[${t - 1}]`, s = t === 1 ? "coords" : e.slice(0, -1).map((r, c) => `coords[${c}]`).join(", ");
+    const t = e.length, n = t === 1 ? "int" : `ivec${Math.min(t, 4)}`, a = t === 1 ? "coords" : `coords[${t - 1}]`, r = t === 1 ? "coords" : e.slice(0, -1).map((s, c) => `coords[${c}]`).join(", ");
     this.userCode = `
         void main() {
             float m = getMomentsAtOutCoords();
             ${n} coords = getOutputCoords();
-            float g = getGradient(${s});
+
+            // Add gradient clipping here
+            float g = clamp(getGradient(${r}), -1.0, 1.0);
             int which = ${a};
 
             float beta = which == 0 ? beta1 : beta2;
@@ -27,8 +29,8 @@ class i {
   }
 }
 function d(o) {
-  const { moments: e, gradient: t } = o.inputs, { beta1: n, beta2: a } = o.attrs, s = o.backend, r = new i(e.shape);
-  return s.runWebGLProgram(r, [e, t], "float32", [[n], [a]]);
+  const { moments: e, gradient: t } = o.inputs, { beta1: n, beta2: a } = o.attrs, r = o.backend, s = new i(e.shape);
+  return r.runWebGLProgram(s, [e, t], "float32", [[n], [a]]);
 }
 const g = {
   kernelName: "AdamMoments",

package/dist/ops/webgl/appendCache.js

@@ -1,4 +1,4 @@
-import { f as p } from "../../index-UdZhlibC.js";
+import { f as p } from "../../index-BoWRt-10.js";
 class m {
   variableNames = ["cache", "item"];
   outputShape;

package/dist/ops/webgl/attentionMask.js

@@ -1,4 +1,4 @@
-import { f as m } from "../../index-UdZhlibC.js";
+import { f as m } from "../../index-BoWRt-10.js";
 class h {
   variableNames = ["q", "k"];
   outputShape;

package/dist/ops/webgl/fusedSoftmax.js

@@ -1,7 +1,7 @@
-import { m as b, s as I, r as k } from "../../RealDiv-CVYNbZxu.js";
-import { r as v } from "../../Reshape-CEsEp0AI.js";
-import { f as w, p as P } from "../../index-UdZhlibC.js";
-import { e as S } from "../../axis_util-5DTW2tFV.js";
+import { m as b, s as I, r as k } from "../../RealDiv-Dy0p8Bvo.js";
+import { r as v } from "../../Reshape-DH5srBP0.js";
+import { f as w, p as P } from "../../index-BoWRt-10.js";
+import { e as S } from "../../axis_util-BzbKo31C.js";
 class T {
   variableNames = ["logits", "maxLogits"];
   outputShape;

package/dist/ops/webgl/gatherSub.js

@@ -1,4 +1,4 @@
-import { f as l } from "../../index-UdZhlibC.js";
+import { f as l } from "../../index-BoWRt-10.js";
 class u {
   variableNames = ["labels", "logits", "values"];
   outputShape;

package/dist/ops/webgl/gelu.js

@@ -1,5 +1,5 @@
-import { f as a } from "../../index-UdZhlibC.js";
-import { u as s, C as x } from "../../kernel_funcs_utils-CXDy3EN7.js";
+import { f as a } from "../../index-BoWRt-10.js";
+import { u as s, C as x } from "../../kernel_funcs_utils-BYKWV8Aa.js";
 const t = 0.7978845608028654, r = 0.044715, c = x + `
     float x3 = x * x * x;
     float inner = x + ${r} * x3;

package/dist/ops/webgl/log.js

@@ -1,6 +1,6 @@
-import { f as e, af as r } from "../../index-UdZhlibC.js";
-import { u as s, l as N } from "../../kernel_funcs_utils-CXDy3EN7.js";
-import { y as l } from "../../shared-DS5waSIY.js";
+import { f as e, af as r } from "../../index-BoWRt-10.js";
+import { u as s, l as N } from "../../kernel_funcs_utils-BYKWV8Aa.js";
+import { y as l } from "../../shared-wS99K7_n.js";
 /**
  * @license
  * Copyright 2020 Google LLC. All Rights Reserved.

package/dist/ops/webgl/matMulGelu.js

@@ -1,7 +1,7 @@
-import { f as _, t as R, e as C, j as A, k as N, l as H, u as O } from "../../index-UdZhlibC.js";
-import { r as f } from "../../Reshape-CEsEp0AI.js";
-import { M as U } from "../../mulmat_packed_gpu-IGPBp6h9.js";
-import { m as E } from "../../mat_mul-dwmZz69e.js";
+import { f as _, t as R, e as C, j as A, k as N, l as H, u as O } from "../../index-BoWRt-10.js";
+import { r as f } from "../../Reshape-DH5srBP0.js";
+import { M as U } from "../../mulmat_packed_gpu-VSekgsNv.js";
+import { m as E } from "../../mat_mul-8m8pfdcx.js";
 const M = 0.7978845608028654, x = 0.044715, j = `
     vec4 x3 = x * x * x;
     vec4 inner = x + ${x} * x3;

package/dist/ops/webgl/matMulMul.js

@@ -1,4 +1,4 @@
-import { f as u } from "../../index-UdZhlibC.js";
+import { f as u } from "../../index-BoWRt-10.js";
 import { batchMatMulGeluImpl as c } from "./matMulGelu.js";
 const M = `
     return a * b;

package/dist/ops/webgl/mulDropout.js

@@ -1,4 +1,4 @@
-import { f as m } from "../../index-UdZhlibC.js";
+import { f as m } from "../../index-BoWRt-10.js";
 class f {
   variableNames = ["a", "b"];
   outputShape;

package/dist/ops/webgl/normRMS.js

@@ -1,5 +1,5 @@
-import { f as p, e as G } from "../../index-UdZhlibC.js";
-import { s as x } from "../../sum-NLYbiDag.js";
+import { f as p, e as G } from "../../index-BoWRt-10.js";
+import { s as x } from "../../sum-66ew2byf.js";
 class y {
   variableNames = ["x", "meanSquare", "gamma"];
   outputShape;

package/dist/ops/webgl/qkv.js

@@ -1,4 +1,4 @@
-import { f as i } from "../../index-UdZhlibC.js";
+import { f as i } from "../../index-BoWRt-10.js";
 class l {
   variableNames = ["x", "kernel"];
   outputShape;

package/dist/ops/webgl/rope.js

@@ -1,4 +1,4 @@
-import { f as u } from "../../index-UdZhlibC.js";
+import { f as u } from "../../index-BoWRt-10.js";
 class l {
   variableNames = ["x", "sin", "cos"];
   outputShape;

package/dist/ops/webgl/scatterSub.js

@@ -1,4 +1,4 @@
-import { f as i } from "../../index-UdZhlibC.js";
+import { f as i } from "../../index-BoWRt-10.js";
 class u {
   variableNames = ["labels", "softmaxProbs", "dy"];
   outputShape;

package/dist/ops/webgpu/adamAdjust.js

@@ -1,7 +1,7 @@
-import { g as p } from "../../webgpu_program-WaoMq-WD.js";
-import { f as d, c as l } from "../../webgpu_util-DhSeP4b6.js";
-import { f } from "../../index-UdZhlibC.js";
-class c {
+import { g as p } from "../../webgpu_program-PFzf1hAQ.js";
+import { f as d, c as l } from "../../webgpu_util-D____QpY.js";
+import { f, a2 as c } from "../../index-BoWRt-10.js";
+class h {
   variableNames = ["moments", "value"];
   outputShape;
   shaderKey = "AdamAdjust";
@@ -26,7 +26,7 @@ class c {
                 let m2Hat = moments.y * uniforms.invbeta2;
 
                 let invSqrt = inverseSqrt(max(m2Hat, 1e-30));
-                let invDenom = invSqrt / (1.0 + uniforms.epsilon * invSqrt);
+                let invDenom = invSqrt / fma(uniforms.epsilon, invSqrt, 1.0);
                 let adjustedValue = fma(-uniforms.learningRate * m1Hat, invDenom, value);
 
                 setOutputAtIndex(index, adjustedValue);
@@ -35,18 +35,20 @@ class c {
     `;
   }
 }
-function h(t) {
-  const { moments: e, value: a } = t.inputs, { beta1: n, beta2: o, learningRate: i, epsilon: s } = t.attrs, r = t.backend, u = new c(a.shape), m = [
+function v(t) {
+  const { moments: e, value: a } = t.inputs, { beta1: n, beta2: s, learningRate: o, epsilon: i } = t.attrs, r = t.backend;
+  c(e.shape, [...a.shape, 2], "Error in AdamAdjust: ");
+  const u = new h(a.shape), m = [
     { type: "float32", data: [1 / n] },
-    { type: "float32", data: [1 / o] },
-    { type: "float32", data: [i] },
-    { type: "float32", data: [s] }
+    { type: "float32", data: [1 / s] },
+    { type: "float32", data: [o] },
+    { type: "float32", data: [i] }
   ];
   return r.runWebGPUProgram(u, [e, a], "float32", m);
 }
-const v = {
+const b = {
   kernelName: "AdamAdjust",
   backendName: "webgpu",
-  kernelFunc: h
+  kernelFunc: v
 };
-f(v);
+f(b);

package/dist/ops/webgpu/adamMoments.js

@@ -1,7 +1,7 @@
-import { g as m } from "../../webgpu_program-WaoMq-WD.js";
-import { f as u, c as p } from "../../webgpu_util-DhSeP4b6.js";
-import { f as c } from "../../index-UdZhlibC.js";
-class d {
+import { g as m } from "../../webgpu_program-PFzf1hAQ.js";
+import { f as u, c as p } from "../../webgpu_util-D____QpY.js";
+import { f as d, a2 as c } from "../../index-BoWRt-10.js";
+class h {
   variableNames = ["moments", "gradient"];
   outputShape;
   shaderKey = "AdamMoments";
@@ -25,7 +25,9 @@ class d {
         ${m("index")} {
             if (index < uniforms.size) {
                 let m: vec2<f32> = moments[index];
-                let g: f32 = gradient[index];
+
+                // Add gradient clipping here
+                let g: f32 = clamp(gradient[index], -1.0, 1.0);
 
                 let newM1 = fma(m.x, uniforms.beta1, g * (1.0 - uniforms.beta1));
                 let newM2 = fma(m.y, uniforms.beta2, g * g * (1.0 - uniforms.beta2));
@@ -37,15 +39,20 @@ class d {
   }
 }
 function f(t) {
-  const { moments: e, gradient: a } = t.inputs, { beta1: n, beta2: o } = t.attrs, s = t.backend, i = new d(e.shape), r = [
-    { type: "float32", data: [n] },
-    { type: "float32", data: [o] }
+  const { moments: e, gradient: o } = t.inputs, { beta1: a, beta2: n } = t.attrs, r = t.backend;
+  if (c(e.shape, [...o.shape, 2], "Error in AdamMoments: "), a < 0 || a >= 1)
+    throw new Error(`Invalid beta1 value: ${a}. Must be in the range [0, 1).`);
+  if (n < 0 || n >= 1)
+    throw new Error(`Invalid beta2 value: ${n}. Must be in the range [0, 1).`);
+  const s = new h(e.shape), i = [
+    { type: "float32", data: [a] },
+    { type: "float32", data: [n] }
   ];
-  return s.runWebGPUProgram(i, [e, a], "float32", r);
+  return r.runWebGPUProgram(s, [e, o], "float32", i);
 }
-const h = {
+const l = {
   kernelName: "AdamMoments",
   backendName: "webgpu",
   kernelFunc: f
 };
-c(h);
+d(l);

package/dist/ops/webgpu/appendCache.js

@@ -1,7 +1,7 @@
-import { g as u } from "../../webgpu_program-WaoMq-WD.js";
-import { f as d, c as m } from "../../webgpu_util-DhSeP4b6.js";
-import { f as l } from "../../index-UdZhlibC.js";
-class f {
+import { g as u } from "../../webgpu_program-PFzf1hAQ.js";
+import { f as d, c as m } from "../../webgpu_util-D____QpY.js";
+import { f as l, a2 as f } from "../../index-BoWRt-10.js";
+class g {
   variableNames = ["cache", "item"];
   outputShape;
   shaderKey = "AppendCache";
@@ -10,12 +10,12 @@ class f {
   workgroupSize = [64, 1, 1];
   size = !0;
   uniforms = "cacheT: i32";
-  constructor(t, a, o, c, i) {
-    const s = Math.min(o + 1, i);
-    this.shaderKey = `AppendCache_${s}`, this.outputShape = [t, a, s, c], this.dispatchLayout = d(this.outputShape), this.dispatch = m(this.dispatchLayout, this.outputShape, this.workgroupSize);
+  constructor(e, t, s, a, c) {
+    const o = Math.min(s + 1, c);
+    this.shaderKey = `AppendCache_${o}`, this.outputShape = [e, t, o, a], this.dispatchLayout = d(this.outputShape), this.dispatch = m(this.dispatchLayout, this.outputShape, this.workgroupSize);
   }
   getUserCode() {
-    const t = this.outputShape[2];
+    const e = this.outputShape[2];
     return `
         ${u("index")} {
             if (index < uniforms.size) {
@@ -26,7 +26,7 @@ class f {
                 let d = coords[3];
 
                 let itemT = 1;
-                let maxSize = ${t};
+                let maxSize = ${e};
                 let totalT = uniforms.cacheT + itemT;
                 let start = select(0, 1, totalT >= maxSize);
 
@@ -45,13 +45,16 @@ class f {
         `;
   }
 }
-function T(e) {
-  const { cache: t, item: a } = e.inputs, { maxSize: o, pastLen: c } = e.attrs, i = e.backend, s = t.shape[0], r = t.shape[2], n = t.shape[1], h = new f(s, n, r, a.shape[3], o), p = [{ type: "int32", data: [c] }];
-  return i.runWebGPUProgram(h, [t, a], "float32", p);
+function S(r) {
+  const { cache: e, item: t } = r.inputs, { maxSize: s, pastLen: a } = r.attrs, c = r.backend, o = e.shape[0], n = e.shape[2], i = e.shape[1];
+  if (f(t.shape, [o, i, 1, t.shape[3]], "Error in AppendCache: "), a < 0 || a > s)
+    throw new Error(`Invalid pastLen value: ${a}. Must be in the range [0, ${s}].`);
+  const h = new g(o, i, n, t.shape[3], s), p = [{ type: "int32", data: [a] }];
+  return c.runWebGPUProgram(h, [e, t], "float32", p);
 }
-const g = {
+const T = {
   kernelName: "AppendCache",
   backendName: "webgpu",
-  kernelFunc: T
+  kernelFunc: S
 };
-l(g);
+l(T);