deepbox 0.1.0 → 0.2.0

package/dist/{chunk-QERHVCHC.cjs → chunk-EUZHJDZ6.cjs}

@@ -1,11 +1,11 @@
 'use strict';
 
-var chunk6AE5FKKQ_cjs = require('./chunk-6AE5FKKQ.cjs');
-var chunkJSCDE774_cjs = require('./chunk-JSCDE774.cjs');
+var chunk724CXHFH_cjs = require('./chunk-724CXHFH.cjs');
+var chunkZ6BGACIH_cjs = require('./chunk-Z6BGACIH.cjs');
 
 // src/nn/index.ts
 var nn_exports = {};
-chunkJSCDE774_cjs.__export(nn_exports, {
+chunkZ6BGACIH_cjs.__export(nn_exports, {
   AvgPool2d: () => AvgPool2d,
   BatchNorm1d: () => BatchNorm1d,
   Conv1d: () => Conv1d,
@@ -53,7 +53,7 @@ function sizeFromShape(shape, context) {
   let size = 1;
   for (const dim of shape) {
     if (!Number.isInteger(dim) || dim < 0) {
-      throw new chunkJSCDE774_cjs.ShapeError(`${context} contains invalid dimension ${String(dim)}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`${context} contains invalid dimension ${String(dim)}`);
     }
     size *= dim;
   }
@@ -71,7 +71,7 @@ function cloneTensorData(t) {
   for (let i = 0; i < data.length; i++) {
     const value = data[i];
     if (value === void 0) {
-      throw new chunkJSCDE774_cjs.DeepboxError("Internal error: tensor data access out of bounds");
+      throw new chunkZ6BGACIH_cjs.DeepboxError("Internal error: tensor data access out of bounds");
     }
     out[i] = value;
   }
@@ -80,35 +80,35 @@ function cloneTensorData(t) {
 function validateStateEntryShape(name, kind, entry) {
   const size = sizeFromShape(entry.shape, `${kind} ${name} shape`);
   if (entry.data.length !== size) {
-    throw new chunkJSCDE774_cjs.ShapeError(
+    throw new chunkZ6BGACIH_cjs.ShapeError(
       `${kind} ${name} data length ${entry.data.length} does not match shape size ${size}`
     );
   }
 }
 function copyStateEntryIntoTensor(name, kind, target, entry) {
   if (!shapesEqual(target.shape, entry.shape)) {
-    throw new chunkJSCDE774_cjs.ShapeError(
+    throw new chunkZ6BGACIH_cjs.ShapeError(
       `${kind} ${name} shape mismatch: expected [${target.shape.join(", ")}], got [${entry.shape.join(", ")}]`
     );
   }
   if (target.dtype !== entry.dtype) {
-    throw new chunkJSCDE774_cjs.DTypeError(
+    throw new chunkZ6BGACIH_cjs.DTypeError(
       `${kind} ${name} dtype mismatch: expected ${target.dtype}, got ${entry.dtype}`
     );
   }
   const size = sizeFromShape(entry.shape, `${kind} ${name} shape`);
-  const logicalStrides = chunk6AE5FKKQ_cjs.computeStrides(target.shape);
+  const logicalStrides = chunk724CXHFH_cjs.computeStrides(target.shape);
   const data = target.data;
   if (target.dtype === "string") {
     if (!Array.isArray(data)) {
-      throw new chunkJSCDE774_cjs.DTypeError(`${kind} ${name} expected string data`);
+      throw new chunkZ6BGACIH_cjs.DTypeError(`${kind} ${name} expected string data`);
     }
     for (let i = 0; i < size; i++) {
       const value = entry.data[i];
       if (typeof value !== "string") {
-        throw new chunkJSCDE774_cjs.DTypeError(`${kind} ${name} expects string data`);
+        throw new chunkZ6BGACIH_cjs.DTypeError(`${kind} ${name} expects string data`);
       }
-      const offset = chunk6AE5FKKQ_cjs.offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);
+      const offset = chunk724CXHFH_cjs.offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);
       data[offset] = value;
     }
     return;
@@ -117,22 +117,22 @@ function copyStateEntryIntoTensor(name, kind, target, entry) {
     for (let i = 0; i < size; i++) {
       const value = entry.data[i];
       if (typeof value !== "bigint") {
-        throw new chunkJSCDE774_cjs.DTypeError(`${kind} ${name} expects bigint data`);
+        throw new chunkZ6BGACIH_cjs.DTypeError(`${kind} ${name} expects bigint data`);
       }
-      const offset = chunk6AE5FKKQ_cjs.offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);
+      const offset = chunk724CXHFH_cjs.offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);
       data[offset] = value;
     }
     return;
   }
   if (Array.isArray(data)) {
-    throw new chunkJSCDE774_cjs.DTypeError(`${kind} ${name} expected numeric data`);
+    throw new chunkZ6BGACIH_cjs.DTypeError(`${kind} ${name} expected numeric data`);
   }
   for (let i = 0; i < size; i++) {
     const value = entry.data[i];
     if (typeof value !== "number") {
-      throw new chunkJSCDE774_cjs.DTypeError(`${kind} ${name} expects numeric data`);
+      throw new chunkZ6BGACIH_cjs.DTypeError(`${kind} ${name} expects numeric data`);
     }
-    const offset = chunk6AE5FKKQ_cjs.offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);
+    const offset = chunk724CXHFH_cjs.offsetFromFlatIndex(i, logicalStrides, target.strides, target.offset);
     data[offset] = value;
   }
 }
@@ -413,7 +413,7 @@ var Module = class _Module {
       const resolved = this.resolveModuleAndName(name);
       if (!resolved) {
         if (providedNames) {
-          throw new chunkJSCDE774_cjs.InvalidParameterError(`Unknown parameter name: ${name}`, "names", name);
+          throw new chunkZ6BGACIH_cjs.InvalidParameterError(`Unknown parameter name: ${name}`, "names", name);
         }
         continue;
       }
@@ -421,11 +421,11 @@ var Module = class _Module {
       const param = module._parameters.get(localName);
       if (!param) {
         if (providedNames) {
-          throw new chunkJSCDE774_cjs.InvalidParameterError(`Unknown parameter name: ${name}`, "names", name);
+          throw new chunkZ6BGACIH_cjs.InvalidParameterError(`Unknown parameter name: ${name}`, "names", name);
         }
         continue;
       }
-      const nextParam = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(param.tensor, { requiresGrad });
+      const nextParam = chunk724CXHFH_cjs.GradTensor.fromTensor(param.tensor, { requiresGrad });
       module._parameters.set(localName, nextParam);
       for (const [key, value] of Object.entries(module)) {
         if (value === param) {
@@ -448,7 +448,7 @@ var Module = class _Module {
   }
   static setTensorDeviceMetadata(target, device) {
     if (!Reflect.set(target, "device", device)) {
-      throw new chunkJSCDE774_cjs.DeepboxError("Failed to update tensor device metadata");
+      throw new chunkZ6BGACIH_cjs.DeepboxError("Failed to update tensor device metadata");
     }
   }
   /**
@@ -486,17 +486,17 @@ var Module = class _Module {
     const namedBuffs = new Map(this.namedBuffers());
     for (const name of namedParams.keys()) {
       if (!(name in parameters)) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(`missing parameter: ${name}`, "stateDict.parameters", name);
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(`missing parameter: ${name}`, "stateDict.parameters", name);
       }
     }
     for (const name of namedBuffs.keys()) {
       if (!(name in buffers)) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(`missing buffer: ${name}`, "stateDict.buffers", name);
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(`missing buffer: ${name}`, "stateDict.buffers", name);
       }
     }
     for (const name of Object.keys(parameters)) {
       if (!namedParams.has(name)) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(
           `unexpected parameter: ${name}`,
           "stateDict.parameters",
           name
@@ -505,7 +505,7 @@ var Module = class _Module {
     }
     for (const name of Object.keys(buffers)) {
       if (!namedBuffs.has(name)) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(`unexpected buffer: ${name}`, "stateDict.buffers", name);
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(`unexpected buffer: ${name}`, "stateDict.buffers", name);
       }
     }
     for (const [name, entry] of Object.entries(parameters)) {
@@ -542,8 +542,8 @@ var Module = class _Module {
    * ```
    */
   to(device) {
-    if (!chunkJSCDE774_cjs.isDevice(device)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("device must be one of: cpu, webgpu, wasm", "device", device);
+    if (!chunkZ6BGACIH_cjs.isDevice(device)) {
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("device must be one of: cpu, webgpu, wasm", "device", device);
     }
     for (const param of this.parameters()) {
       _Module.setTensorDeviceMetadata(param.tensor, device);
@@ -613,7 +613,7 @@ var Sequential = class extends Module {
   constructor(...layers) {
     super();
     if (layers.length === 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "Sequential requires at least one layer",
         "layers",
         layers.length
@@ -623,7 +623,7 @@ var Sequential = class extends Module {
     for (let i = 0; i < layers.length; i++) {
       const layer = layers[i];
       if (!layer) {
-        throw new chunkJSCDE774_cjs.DeepboxError(`Layer at index ${i} is undefined`);
+        throw new chunkZ6BGACIH_cjs.DeepboxError(`Layer at index ${i} is undefined`);
       }
       this.registerModule(String(i), layer);
     }
@@ -640,7 +640,7 @@ var Sequential = class extends Module {
    */
   forward(...inputs) {
     if (inputs.length !== 1) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "Sequential.forward expects a single input tensor",
         "inputs",
         inputs.length
@@ -648,7 +648,7 @@ var Sequential = class extends Module {
     }
     const input = inputs[0];
     if (!input) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "Sequential.forward expects a single input tensor",
         "input",
         input
@@ -658,11 +658,11 @@ var Sequential = class extends Module {
     for (let i = 0; i < this.layers.length; i++) {
       const layer = this.layers[i];
       if (!layer) {
-        throw new chunkJSCDE774_cjs.DeepboxError(`Layer at index ${i} is undefined`);
+        throw new chunkZ6BGACIH_cjs.DeepboxError(`Layer at index ${i} is undefined`);
       }
       const result = layer.call(output);
       if (Array.isArray(result)) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(
           `Sequential does not support layers that return multiple tensors (layer ${i})`,
           "layer",
           i
@@ -682,14 +682,14 @@ var Sequential = class extends Module {
    */
   getLayer(index) {
     if (index < 0 || index >= this.layers.length) {
-      throw new chunkJSCDE774_cjs.IndexError(`Layer index ${index} out of bounds [0, ${this.layers.length})`, {
+      throw new chunkZ6BGACIH_cjs.IndexError(`Layer index ${index} out of bounds [0, ${this.layers.length})`, {
         index,
         validRange: [0, this.layers.length - 1]
       });
     }
     const layer = this.layers[index];
     if (!layer) {
-      throw new chunkJSCDE774_cjs.DeepboxError(`Layer at index ${index} is undefined`);
+      throw new chunkZ6BGACIH_cjs.DeepboxError(`Layer at index ${index} is undefined`);
     }
     return layer;
   }
@@ -731,8 +731,8 @@ var Sequential = class extends Module {
 // src/nn/layers/activations.ts
 var ReLU = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) return input.relu();
-    return chunk6AE5FKKQ_cjs.relu(input);
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) return input.relu();
+    return chunk724CXHFH_cjs.relu(input);
   }
   toString() {
     return "ReLU()";
@@ -740,8 +740,8 @@ var ReLU = class extends Module {
 };
 var Sigmoid = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) return input.sigmoid();
-    return chunk6AE5FKKQ_cjs.sigmoid(input);
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) return input.sigmoid();
+    return chunk724CXHFH_cjs.sigmoid(input);
   }
   toString() {
     return "Sigmoid()";
@@ -749,8 +749,8 @@ var Sigmoid = class extends Module {
 };
 var Tanh = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) return input.tanh();
-    return chunk6AE5FKKQ_cjs.tanh(input);
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) return input.tanh();
+    return chunk724CXHFH_cjs.tanh(input);
   }
   toString() {
     return "Tanh()";
@@ -763,8 +763,8 @@ var LeakyReLU = class extends Module {
     this.alpha = alpha;
   }
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) return input.leakyRelu(this.alpha);
-    return chunk6AE5FKKQ_cjs.leakyRelu(input, this.alpha);
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) return input.leakyRelu(this.alpha);
+    return chunk724CXHFH_cjs.leakyRelu(input, this.alpha);
   }
   toString() {
     return `LeakyReLU(alpha=${this.alpha})`;
@@ -777,8 +777,8 @@ var ELU = class extends Module {
     this.alpha = alpha;
   }
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) return input.elu(this.alpha);
-    return chunk6AE5FKKQ_cjs.elu(input, this.alpha);
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) return input.elu(this.alpha);
+    return chunk724CXHFH_cjs.elu(input, this.alpha);
   }
   toString() {
     return `ELU(alpha=${this.alpha})`;
@@ -786,8 +786,8 @@ var ELU = class extends Module {
 };
 var GELU = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) return input.gelu();
-    return chunk6AE5FKKQ_cjs.gelu(input);
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) return input.gelu();
+    return chunk724CXHFH_cjs.gelu(input);
   }
   toString() {
     return "GELU()";
@@ -800,10 +800,10 @@ var Softmax = class extends Module {
     this.axis = axis;
   }
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) {
-      return chunk6AE5FKKQ_cjs.softmax2(input, chunkJSCDE774_cjs.normalizeAxis(this.axis, input.tensor.ndim));
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
+      return chunk724CXHFH_cjs.softmax2(input, chunkZ6BGACIH_cjs.normalizeAxis(this.axis, input.tensor.ndim));
     }
-    return chunk6AE5FKKQ_cjs.softmax(input, this.axis);
+    return chunk724CXHFH_cjs.softmax(input, this.axis);
   }
   toString() {
     return `Softmax(axis=${this.axis})`;
@@ -816,10 +816,10 @@ var LogSoftmax = class extends Module {
     this.axis = axis;
   }
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) {
-      return chunk6AE5FKKQ_cjs.logSoftmax2(input, chunkJSCDE774_cjs.normalizeAxis(this.axis, input.tensor.ndim));
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
+      return chunk724CXHFH_cjs.logSoftmax2(input, chunkZ6BGACIH_cjs.normalizeAxis(this.axis, input.tensor.ndim));
     }
-    return chunk6AE5FKKQ_cjs.logSoftmax(input, this.axis);
+    return chunk724CXHFH_cjs.logSoftmax(input, this.axis);
   }
   toString() {
     return `LogSoftmax(axis=${this.axis})`;
@@ -827,12 +827,13 @@ var LogSoftmax = class extends Module {
 };
 var Softplus = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) {
-      return chunk6AE5FKKQ_cjs.GradTensor.fromTensor(chunk6AE5FKKQ_cjs.softplus(input.tensor), {
-        requiresGrad: false
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
+      const one = chunk724CXHFH_cjs.GradTensor.scalar(1, {
+        dtype: input.dtype === "float64" ? "float64" : "float32"
       });
+      return one.add(input.exp()).log();
     }
-    return chunk6AE5FKKQ_cjs.softplus(input);
+    return chunk724CXHFH_cjs.softplus(input);
   }
   toString() {
     return "Softplus()";
@@ -840,12 +841,10 @@ var Softplus = class extends Module {
 };
 var Swish = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) {
-      return chunk6AE5FKKQ_cjs.GradTensor.fromTensor(chunk6AE5FKKQ_cjs.swish(input.tensor), {
-        requiresGrad: false
-      });
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
+      return input.mul(input.sigmoid());
     }
-    return chunk6AE5FKKQ_cjs.swish(input);
+    return chunk724CXHFH_cjs.swish(input);
   }
   toString() {
     return "Swish()";
@@ -853,12 +852,14 @@ var Swish = class extends Module {
 };
 var Mish = class extends Module {
   forward(input) {
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) {
-      return chunk6AE5FKKQ_cjs.GradTensor.fromTensor(chunk6AE5FKKQ_cjs.mish(input.tensor), {
-        requiresGrad: false
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
+      const one = chunk724CXHFH_cjs.GradTensor.scalar(1, {
+        dtype: input.dtype === "float64" ? "float64" : "float32"
       });
+      const sp = one.add(input.exp()).log();
+      return input.mul(sp.tanh());
     }
-    return chunk6AE5FKKQ_cjs.mish(input);
+    return chunk724CXHFH_cjs.mish(input);
   }
   toString() {
     return "Mish()";
@@ -878,7 +879,7 @@ var Dropout = class extends Module {
   constructor(p = 0.5) {
     super();
     if (!Number.isFinite(p) || p < 0 || p >= 1) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(`Dropout probability must be in [0, 1), got ${p}`, "p", p);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(`Dropout probability must be in [0, 1), got ${p}`, "p", p);
     }
     this.p = p;
   }
@@ -889,11 +890,11 @@ var Dropout = class extends Module {
    * @returns Output tensor with same shape as input
    */
   forward(input) {
-    const inputTensor = input instanceof chunk6AE5FKKQ_cjs.GradTensor ? input : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(input);
+    const inputTensor = chunk724CXHFH_cjs.GradTensor.isGradTensor(input) ? input : chunk724CXHFH_cjs.GradTensor.fromTensor(input);
     if (inputTensor.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("Dropout does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("Dropout does not support string dtype");
     }
-    return chunk6AE5FKKQ_cjs.dropout(inputTensor, this.p, this.training);
+    return chunk724CXHFH_cjs.dropout(inputTensor, this.p, this.training);
   }
   /**
    * Get string representation of the layer.
@@ -938,14 +939,14 @@ var Linear = class extends Module {
   constructor(inFeatures, outFeatures, options = {}) {
     super();
     if (inFeatures <= 0 || !Number.isInteger(inFeatures)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "inFeatures must be a positive integer",
         "inFeatures",
         inFeatures
       );
     }
     if (outFeatures <= 0 || !Number.isInteger(outFeatures)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "outFeatures must be a positive integer",
         "outFeatures",
         outFeatures
@@ -955,42 +956,52 @@ var Linear = class extends Module {
     this.outFeatures = outFeatures;
     this.useBias = options.bias ?? true;
     const stdDev = Math.sqrt(2 / inFeatures);
-    const weightTensor = chunk6AE5FKKQ_cjs.randn([outFeatures, inFeatures], {
+    const weightTensor = chunk724CXHFH_cjs.randn([outFeatures, inFeatures], {
       dtype: options.dtype ?? "float32",
       device: options.device ?? "cpu"
     });
-    const scaledWeight = chunk6AE5FKKQ_cjs.mulScalar(weightTensor, stdDev);
-    this.weightParam = chunk6AE5FKKQ_cjs.parameter(scaledWeight);
+    const scaledWeight = chunk724CXHFH_cjs.mulScalar(weightTensor, stdDev);
+    this.weightParam = chunk724CXHFH_cjs.parameter(scaledWeight);
     this.weight = this.weightParam.tensor;
     this.registerParameter("weight", this.weightParam);
     if (this.useBias) {
-      const biasTensor = chunk6AE5FKKQ_cjs.zeros([outFeatures], {
+      const biasTensor = chunk724CXHFH_cjs.zeros([outFeatures], {
         dtype: options.dtype ?? "float32",
         device: options.device ?? "cpu"
       });
-      this.biasParam = chunk6AE5FKKQ_cjs.parameter(biasTensor);
+      this.biasParam = chunk724CXHFH_cjs.parameter(biasTensor);
       this.bias = this.biasParam.tensor;
       this.registerParameter("bias", this.biasParam);
     }
   }
   forward(input) {
-    const inputTensor = input instanceof chunk6AE5FKKQ_cjs.GradTensor ? input.tensor : input;
+    let inputTensor = chunk724CXHFH_cjs.GradTensor.isGradTensor(input) ? input.tensor : input;
     if (inputTensor.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("Linear layer does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("Linear layer does not support string dtype");
+    }
+    if (inputTensor.dtype !== this.weight.dtype && inputTensor.dtype !== "int64") {
+      const castData = new Float32Array(
+        inputTensor.data
+      );
+      const castTensor = chunk724CXHFH_cjs.reshape(chunk724CXHFH_cjs.tensor(castData), inputTensor.shape);
+      inputTensor = castTensor;
+      if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
+        input = chunk724CXHFH_cjs.parameter(castTensor);
+      }
     }
     if (inputTensor.ndim < 1) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Linear layer expects at least 1D input; got ndim=${inputTensor.ndim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Linear layer expects at least 1D input; got ndim=${inputTensor.ndim}`);
     }
     const inputFeatures = inputTensor.shape[inputTensor.shape.length - 1] ?? 0;
     if (inputFeatures !== this.inFeatures) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `Linear layer expects ${this.inFeatures} input features; got ${inputFeatures}`
       );
     }
     const isVectorInput = inputTensor.ndim === 1;
     const batchSize = inputTensor.size / this.inFeatures;
     const outputShape = isVectorInput ? [this.outFeatures] : [...inputTensor.shape.slice(0, -1), this.outFeatures];
-    if (input instanceof chunk6AE5FKKQ_cjs.GradTensor) {
+    if (chunk724CXHFH_cjs.GradTensor.isGradTensor(input)) {
       const input2d2 = input.reshape([batchSize, this.inFeatures]);
       const output2d2 = input2d2.matmul(this.weightParam.transpose());
       let output2 = output2d2.reshape(outputShape);
@@ -999,11 +1010,11 @@ var Linear = class extends Module {
       }
       return output2;
     }
-    const input2d = chunk6AE5FKKQ_cjs.reshape(inputTensor, [batchSize, this.inFeatures]);
-    const output2d = chunk6AE5FKKQ_cjs.dot(input2d, chunk6AE5FKKQ_cjs.transpose(this.weight));
-    const output = chunk6AE5FKKQ_cjs.reshape(output2d, outputShape);
+    const input2d = chunk724CXHFH_cjs.reshape(inputTensor, [batchSize, this.inFeatures]);
+    const output2d = chunk724CXHFH_cjs.dot(input2d, chunk724CXHFH_cjs.transpose(this.weight));
+    const output = chunk724CXHFH_cjs.reshape(output2d, outputShape);
     if (this.useBias && this.bias) {
-      return chunk6AE5FKKQ_cjs.add(output, this.bias);
+      return chunk724CXHFH_cjs.add(output, this.bias);
     }
     return output;
   }
@@ -1048,37 +1059,37 @@ var Linear = class extends Module {
 
 // src/nn/layers/normalization.ts
 function toContiguousTensor(t) {
-  if (chunk6AE5FKKQ_cjs.isContiguous(t.shape, t.strides)) {
+  if (chunk724CXHFH_cjs.isContiguous(t.shape, t.strides)) {
     return t;
   }
   if (t.dtype === "string") {
-    throw new chunkJSCDE774_cjs.DTypeError("Normalization does not support string dtype");
+    throw new chunkZ6BGACIH_cjs.DTypeError("Normalization does not support string dtype");
   }
-  const Ctor = chunkJSCDE774_cjs.dtypeToTypedArrayCtor(t.dtype);
+  const Ctor = chunkZ6BGACIH_cjs.dtypeToTypedArrayCtor(t.dtype);
   const out = new Ctor(t.size);
-  const logicalStrides = chunk6AE5FKKQ_cjs.computeStrides(t.shape);
+  const logicalStrides = chunk724CXHFH_cjs.computeStrides(t.shape);
   const data = t.data;
   if (Array.isArray(data)) {
-    throw new chunkJSCDE774_cjs.DTypeError("Normalization does not support string dtype");
+    throw new chunkZ6BGACIH_cjs.DTypeError("Normalization does not support string dtype");
   }
   if (data instanceof BigInt64Array) {
     if (!(out instanceof BigInt64Array)) {
-      throw new chunkJSCDE774_cjs.DTypeError("Expected int64 output buffer for int64 tensor");
+      throw new chunkZ6BGACIH_cjs.DTypeError("Expected int64 output buffer for int64 tensor");
     }
     for (let i = 0; i < t.size; i++) {
-      const offset = chunk6AE5FKKQ_cjs.offsetFromFlatIndex(i, logicalStrides, t.strides, t.offset);
-      out[i] = chunkJSCDE774_cjs.getBigIntElement(data, offset);
+      const offset = chunk724CXHFH_cjs.offsetFromFlatIndex(i, logicalStrides, t.strides, t.offset);
+      out[i] = chunkZ6BGACIH_cjs.getBigIntElement(data, offset);
     }
   } else {
     if (out instanceof BigInt64Array) {
-      throw new chunkJSCDE774_cjs.DTypeError("Unexpected int64 output buffer for numeric tensor");
+      throw new chunkZ6BGACIH_cjs.DTypeError("Unexpected int64 output buffer for numeric tensor");
     }
     for (let i = 0; i < t.size; i++) {
-      const offset = chunk6AE5FKKQ_cjs.offsetFromFlatIndex(i, logicalStrides, t.strides, t.offset);
-      out[i] = chunkJSCDE774_cjs.getNumericElement(data, offset);
+      const offset = chunk724CXHFH_cjs.offsetFromFlatIndex(i, logicalStrides, t.strides, t.offset);
+      out[i] = chunkZ6BGACIH_cjs.getNumericElement(data, offset);
     }
   }
-  return chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+  return chunk724CXHFH_cjs.Tensor.fromTypedArray({
     data: out,
     shape: t.shape,
     dtype: t.dtype,
@@ -1098,7 +1109,7 @@ var BatchNorm1d = class extends Module {
   constructor(numFeatures, options = {}) {
     super();
     if (!Number.isFinite(numFeatures) || numFeatures <= 0 || Math.trunc(numFeatures) !== numFeatures) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "numFeatures must be a positive integer",
         "numFeatures",
         numFeatures
@@ -1107,11 +1118,11 @@ var BatchNorm1d = class extends Module {
     this.numFeatures = numFeatures;
     this.eps = options.eps ?? 1e-5;
     if (!Number.isFinite(this.eps) || this.eps <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("eps must be a positive number", "eps", this.eps);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("eps must be a positive number", "eps", this.eps);
     }
     this.momentum = options.momentum ?? 0.1;
     if (!Number.isFinite(this.momentum) || this.momentum < 0 || this.momentum > 1) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "momentum must be in range [0, 1]",
         "momentum",
         this.momentum
@@ -1120,17 +1131,17 @@ var BatchNorm1d = class extends Module {
     this.affine = options.affine ?? true;
     this.trackRunningStats = options.trackRunningStats ?? true;
     if (this.affine) {
-      const gamma = chunk6AE5FKKQ_cjs.ones([numFeatures]);
-      const beta = chunk6AE5FKKQ_cjs.zeros([numFeatures]);
-      this.gamma = chunk6AE5FKKQ_cjs.parameter(gamma);
-      this.beta = chunk6AE5FKKQ_cjs.parameter(beta);
+      const gamma = chunk724CXHFH_cjs.ones([numFeatures]);
+      const beta = chunk724CXHFH_cjs.zeros([numFeatures]);
+      this.gamma = chunk724CXHFH_cjs.parameter(gamma);
+      this.beta = chunk724CXHFH_cjs.parameter(beta);
       this.registerParameter("weight", this.gamma);
       this.registerParameter("bias", this.beta);
     }
-    this.runningMean = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(chunk6AE5FKKQ_cjs.zeros([numFeatures]), {
+    this.runningMean = chunk724CXHFH_cjs.GradTensor.fromTensor(chunk724CXHFH_cjs.zeros([numFeatures]), {
       requiresGrad: false
     });
-    this.runningVar = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(chunk6AE5FKKQ_cjs.ones([numFeatures]), {
+    this.runningVar = chunk724CXHFH_cjs.GradTensor.fromTensor(chunk724CXHFH_cjs.ones([numFeatures]), {
       requiresGrad: false
     });
     if (this.trackRunningStats) {
@@ -1139,17 +1150,17 @@ var BatchNorm1d = class extends Module {
     }
   }
   forward(x) {
-    const input = x instanceof chunk6AE5FKKQ_cjs.GradTensor ? x : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(x);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(x) ? x : chunk724CXHFH_cjs.GradTensor.fromTensor(x);
     const inputDtype = input.dtype;
     if (inputDtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("BatchNorm1d does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("BatchNorm1d does not support string dtype");
     }
     if (input.ndim !== 2 && input.ndim !== 3) {
-      throw new chunkJSCDE774_cjs.ShapeError(`BatchNorm1d expects 2D or 3D input; got ndim=${input.ndim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`BatchNorm1d expects 2D or 3D input; got ndim=${input.ndim}`);
     }
     const nFeatures = input.shape[1] ?? 0;
     if (nFeatures !== this.numFeatures) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected ${this.numFeatures} features, got ${nFeatures}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected ${this.numFeatures} features, got ${nFeatures}`);
     }
     const useBatchStats = this.training || !this.trackRunningStats;
     let mean2;
@@ -1159,36 +1170,36 @@ var BatchNorm1d = class extends Module {
       const batch = input.shape[0] ?? 0;
       const length = input.shape[2] ?? 0;
       const flat = batch * length;
-      const numericInputDtype = chunkJSCDE774_cjs.ensureNumericDType(inputDtype, "BatchNorm1d");
-      inputReshaped = input.transpose([0, 2, 1]).mul(chunk6AE5FKKQ_cjs.GradTensor.scalar(1, { dtype: numericInputDtype })).reshape([flat, nFeatures]);
+      const numericInputDtype = chunkZ6BGACIH_cjs.ensureNumericDType(inputDtype, "BatchNorm1d");
+      inputReshaped = input.transpose([0, 2, 1]).mul(chunk724CXHFH_cjs.GradTensor.scalar(1, { dtype: numericInputDtype })).reshape([flat, nFeatures]);
     }
     if (useBatchStats) {
       if (inputReshaped.shape[0] === 0) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(
           "BatchNorm requires at least one element",
           "input",
           input.shape
         );
       }
       mean2 = inputReshaped.mean(0);
-      varVal = chunk6AE5FKKQ_cjs.variance2(inputReshaped, 0, 0);
+      varVal = chunk724CXHFH_cjs.variance2(inputReshaped, 0, 0);
       if (this.trackRunningStats) {
-        chunk6AE5FKKQ_cjs.noGrad(() => {
+        chunk724CXHFH_cjs.noGrad(() => {
           const n = inputReshaped.shape[0] ?? 0;
-          const unbiasedVar = n > 1 ? chunk6AE5FKKQ_cjs.variance2(inputReshaped, 0, 1) : chunk6AE5FKKQ_cjs.variance2(inputReshaped, 0, 0);
+          const unbiasedVar = n > 1 ? chunk724CXHFH_cjs.variance2(inputReshaped, 0, 1) : chunk724CXHFH_cjs.variance2(inputReshaped, 0, 0);
           const m = this.momentum;
           const statsDtype = this.runningMean.dtype;
           if (statsDtype === "string") {
-            throw new chunkJSCDE774_cjs.DTypeError("BatchNorm running statistics must be numeric");
+            throw new chunkZ6BGACIH_cjs.DTypeError("BatchNorm running statistics must be numeric");
           }
-          const oneMinusM = chunk6AE5FKKQ_cjs.GradTensor.scalar(1 - m, { dtype: statsDtype });
-          const mScalar = chunk6AE5FKKQ_cjs.GradTensor.scalar(m, { dtype: statsDtype });
+          const oneMinusM = chunk724CXHFH_cjs.GradTensor.scalar(1 - m, { dtype: statsDtype });
+          const mScalar = chunk724CXHFH_cjs.GradTensor.scalar(m, { dtype: statsDtype });
           const newMean = this.runningMean.mul(oneMinusM).add(mean2.mul(mScalar));
           const newVar = this.runningVar.mul(oneMinusM).add(unbiasedVar.mul(mScalar));
-          this.runningMean = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(newMean.tensor, {
+          this.runningMean = chunk724CXHFH_cjs.GradTensor.fromTensor(newMean.tensor, {
             requiresGrad: false
           });
-          this.runningVar = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(newVar.tensor, {
+          this.runningVar = chunk724CXHFH_cjs.GradTensor.fromTensor(newVar.tensor, {
             requiresGrad: false
           });
           this.registerBuffer("running_mean", this.runningMean.tensor);
@@ -1208,7 +1219,7 @@ var BatchNorm1d = class extends Module {
       meanBroadcast = mean2.reshape([1, nFeatures]);
       varBroadcast = varVal.reshape([1, nFeatures]);
     }
-    const epsTensor = chunk6AE5FKKQ_cjs.GradTensor.scalar(this.eps, { dtype: inputDtype });
+    const epsTensor = chunk724CXHFH_cjs.GradTensor.scalar(this.eps, { dtype: inputDtype });
     const denom = varBroadcast.add(epsTensor).sqrt();
     let out = input.sub(meanBroadcast).div(denom);
     if (this.affine && this.gamma && this.beta) {
@@ -1239,7 +1250,7 @@ var LayerNorm = class extends Module {
     super();
     this.normalizedShape = typeof normalizedShape === "number" ? [normalizedShape] : Array.from(normalizedShape);
     if (this.normalizedShape.length === 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "normalizedShape must contain at least one dimension",
         "normalizedShape",
         normalizedShape
@@ -1247,7 +1258,7 @@ var LayerNorm = class extends Module {
     }
     for (const dim of this.normalizedShape) {
       if (!Number.isFinite(dim) || dim <= 0 || Math.trunc(dim) !== dim) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(
           "All dimensions in normalizedShape must be positive integers",
           "normalizedShape",
           normalizedShape
@@ -1256,38 +1267,38 @@ var LayerNorm = class extends Module {
     }
     this.eps = options.eps ?? 1e-5;
     if (!Number.isFinite(this.eps) || this.eps <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("eps must be a positive number", "eps", this.eps);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("eps must be a positive number", "eps", this.eps);
     }
     this.elementwiseAffine = options.elementwiseAffine ?? true;
     if (this.elementwiseAffine) {
-      this.gamma = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.ones(this.normalizedShape));
-      this.beta = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.zeros(this.normalizedShape));
+      this.gamma = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.ones(this.normalizedShape));
+      this.beta = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.zeros(this.normalizedShape));
       this.registerParameter("weight", this.gamma);
       this.registerParameter("bias", this.beta);
     }
   }
   forward(x) {
-    const input = x instanceof chunk6AE5FKKQ_cjs.GradTensor ? x : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(x);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(x) ? x : chunk724CXHFH_cjs.GradTensor.fromTensor(x);
     const inputDtype = input.dtype;
     if (inputDtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("LayerNorm does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("LayerNorm does not support string dtype");
     }
     let workingInput = input;
-    if (!chunk6AE5FKKQ_cjs.isContiguous(input.tensor.shape, input.tensor.strides)) {
+    if (!chunk724CXHFH_cjs.isContiguous(input.tensor.shape, input.tensor.strides)) {
       const contiguous = toContiguousTensor(input.tensor);
-      workingInput = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(contiguous, {
+      workingInput = chunk724CXHFH_cjs.GradTensor.fromTensor(contiguous, {
         requiresGrad: input.requiresGrad
       });
     }
     const inputShape = workingInput.shape;
     const normShape = this.normalizedShape;
     if (normShape.length > inputShape.length) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Input shape ${inputShape} too small for normalizedShape ${normShape}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Input shape ${inputShape} too small for normalizedShape ${normShape}`);
     }
     const suffixStart = inputShape.length - normShape.length;
     for (let i = 0; i < normShape.length; i++) {
       if (inputShape[suffixStart + i] !== normShape[i]) {
-        throw new chunkJSCDE774_cjs.ShapeError(
+        throw new chunkZ6BGACIH_cjs.ShapeError(
           `Input shape ${inputShape} does not end with normalizedShape ${normShape}`
         );
       }
@@ -1297,9 +1308,9 @@ var LayerNorm = class extends Module {
     const flattenedShape = [...outerDims, normSize];
     const inputReshaped = workingInput.reshape(flattenedShape);
     const mean2 = inputReshaped.mean(-1, true);
-    const varVal = chunk6AE5FKKQ_cjs.variance2(inputReshaped, -1, 0);
+    const varVal = chunk724CXHFH_cjs.variance2(inputReshaped, -1, 0);
     const varReshaped = varVal.reshape(mean2.shape);
-    const epsTensor = chunk6AE5FKKQ_cjs.GradTensor.scalar(this.eps, { dtype: inputDtype });
+    const epsTensor = chunk724CXHFH_cjs.GradTensor.scalar(this.eps, { dtype: inputDtype });
     const denom = varReshaped.add(epsTensor).sqrt();
     const normalizedReshaped = inputReshaped.sub(mean2).div(denom);
     let out = normalizedReshaped.reshape(inputShape);
@@ -1351,13 +1362,13 @@ var MultiheadAttention = class extends Module {
   constructor(embedDim, numHeads, options = {}) {
     super();
     if (!Number.isInteger(embedDim) || embedDim <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("embedDim must be a positive integer", "embedDim", embedDim);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("embedDim must be a positive integer", "embedDim", embedDim);
     }
     if (!Number.isInteger(numHeads) || numHeads <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("numHeads must be a positive integer", "numHeads", numHeads);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("numHeads must be a positive integer", "numHeads", numHeads);
     }
     if (embedDim % numHeads !== 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         `embedDim (${embedDim}) must be divisible by numHeads (${numHeads})`,
         "embedDim",
         embedDim
@@ -1365,7 +1376,7 @@ var MultiheadAttention = class extends Module {
     }
     const dropout2 = options.dropout ?? 0;
     if (!Number.isFinite(dropout2) || dropout2 < 0 || dropout2 >= 1) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("dropout must be in [0, 1)", "dropout", dropout2);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("dropout must be in [0, 1)", "dropout", dropout2);
     }
     this.embedDim = embedDim;
     this.numHeads = numHeads;
@@ -1374,19 +1385,19 @@ var MultiheadAttention = class extends Module {
     this.useBias = options.bias ?? true;
     this.dropout = dropout2;
     const stdDev = Math.sqrt(2 / (embedDim + embedDim));
-    this.wQ = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([embedDim, embedDim]), stdDev));
-    this.wK = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([embedDim, embedDim]), stdDev));
-    this.wV = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([embedDim, embedDim]), stdDev));
-    this.wO = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([embedDim, embedDim]), stdDev));
+    this.wQ = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([embedDim, embedDim]), stdDev));
+    this.wK = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([embedDim, embedDim]), stdDev));
+    this.wV = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([embedDim, embedDim]), stdDev));
+    this.wO = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([embedDim, embedDim]), stdDev));
     this.registerParameter("in_proj_weight_q", this.wQ);
     this.registerParameter("in_proj_weight_k", this.wK);
     this.registerParameter("in_proj_weight_v", this.wV);
     this.registerParameter("out_proj_weight", this.wO);
     if (this.useBias) {
-      this.bQ = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.zeros([embedDim]));
-      this.bK = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.zeros([embedDim]));
-      this.bV = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.zeros([embedDim]));
-      this.bO = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.zeros([embedDim]));
+      this.bQ = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.zeros([embedDim]));
+      this.bK = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.zeros([embedDim]));
+      this.bV = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.zeros([embedDim]));
+      this.bO = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.zeros([embedDim]));
       this.registerParameter("in_proj_bias_q", this.bQ);
       this.registerParameter("in_proj_bias_k", this.bK);
       this.registerParameter("in_proj_bias_v", this.bV);
@@ -1403,7 +1414,7 @@ var MultiheadAttention = class extends Module {
    */
   forward(...inputs) {
     if (inputs.length < 1 || inputs.length > 3) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "MultiheadAttention.forward expects 1 to 3 input tensors",
         "inputs",
         inputs.length
@@ -1411,25 +1422,25 @@ var MultiheadAttention = class extends Module {
     }
     const queryInput = inputs[0];
     if (queryInput === void 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Query tensor is required", "query", queryInput);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Query tensor is required", "query", queryInput);
     }
-    const query = queryInput instanceof chunk6AE5FKKQ_cjs.GradTensor ? queryInput : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(queryInput);
+    const query = chunk724CXHFH_cjs.GradTensor.isGradTensor(queryInput) ? queryInput : chunk724CXHFH_cjs.GradTensor.fromTensor(queryInput);
     const keyInput = inputs[1] ?? queryInput;
-    const key = keyInput instanceof chunk6AE5FKKQ_cjs.GradTensor ? keyInput : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(keyInput);
+    const key = chunk724CXHFH_cjs.GradTensor.isGradTensor(keyInput) ? keyInput : chunk724CXHFH_cjs.GradTensor.fromTensor(keyInput);
     const valueInput = inputs[2] ?? queryInput;
-    const value = valueInput instanceof chunk6AE5FKKQ_cjs.GradTensor ? valueInput : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(valueInput);
-    if (query.dtype === "string") throw new chunkJSCDE774_cjs.DTypeError("String tensors are not supported");
+    const value = chunk724CXHFH_cjs.GradTensor.isGradTensor(valueInput) ? valueInput : chunk724CXHFH_cjs.GradTensor.fromTensor(valueInput);
+    if (query.dtype === "string") throw new chunkZ6BGACIH_cjs.DTypeError("String tensors are not supported");
     if (query.ndim !== key.ndim || query.ndim !== value.ndim) {
-      throw new chunkJSCDE774_cjs.ShapeError("query, key, and value must have same rank");
+      throw new chunkZ6BGACIH_cjs.ShapeError("query, key, and value must have same rank");
     }
     if (query.ndim !== 2 && query.ndim !== 3) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Query must be 2D or 3D; got ndim=${query.ndim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Query must be 2D or 3D; got ndim=${query.ndim}`);
     }
     if (key.ndim !== 2 && key.ndim !== 3) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Key must be 2D or 3D; got ndim=${key.ndim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Key must be 2D or 3D; got ndim=${key.ndim}`);
     }
     if (value.ndim !== 2 && value.ndim !== 3) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Value must be 2D or 3D; got ndim=${value.ndim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Value must be 2D or 3D; got ndim=${value.ndim}`);
     }
     let q = query;
     let k = key;
@@ -1443,21 +1454,21 @@ var MultiheadAttention = class extends Module {
     const seqLenV = v.shape[1] ?? 0;
     const embedDim = q.shape[2] ?? 0;
     if (embedDim !== this.embedDim) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Query embedDim mismatch: expected ${this.embedDim}, got ${embedDim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Query embedDim mismatch: expected ${this.embedDim}, got ${embedDim}`);
     }
     if (k.shape[2] !== this.embedDim) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Key embedDim mismatch: expected ${this.embedDim}, got ${k.shape[2]}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Key embedDim mismatch: expected ${this.embedDim}, got ${k.shape[2]}`);
     }
     if (v.shape[2] !== this.embedDim) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Value embedDim mismatch: expected ${this.embedDim}, got ${v.shape[2]}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Value embedDim mismatch: expected ${this.embedDim}, got ${v.shape[2]}`);
     }
     if (k.shape[0] !== batchSize || v.shape[0] !== batchSize) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `batch size mismatch: query=${batchSize}, key=${k.shape[0]}, value=${v.shape[0]}`
       );
     }
     if (seqLenK !== seqLenV) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Key/value sequence length mismatch: key=${seqLenK}, value=${seqLenV}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Key/value sequence length mismatch: key=${seqLenK}, value=${seqLenV}`);
     }
     let Q = q.matmul(this.wQ.transpose());
     if (this.bQ) Q = Q.add(this.bQ);
@@ -1471,12 +1482,12 @@ var MultiheadAttention = class extends Module {
     K = K.reshape([batchSize, seqLenK, H, D]).transpose([0, 2, 1, 3]);
     V = V.reshape([batchSize, seqLenV, H, D]).transpose([0, 2, 1, 3]);
     let scores = Q.matmul(K.transpose([0, 1, 3, 2]));
-    scores = scores.div(chunk6AE5FKKQ_cjs.GradTensor.scalar(this.scale));
-    let attn = chunk6AE5FKKQ_cjs.softmax2(scores, -1);
-    attn = chunk6AE5FKKQ_cjs.dropout(attn, this.dropout, this.training);
+    scores = scores.div(chunk724CXHFH_cjs.GradTensor.scalar(this.scale));
+    let attn = chunk724CXHFH_cjs.softmax2(scores, -1);
+    attn = chunk724CXHFH_cjs.dropout(attn, this.dropout, this.training);
     const context = attn.matmul(V);
-    const contextDtype = chunkJSCDE774_cjs.ensureNumericDType(context.dtype, "MultiheadAttention");
-    const contextReshaped = context.transpose([0, 2, 1, 3]).mul(chunk6AE5FKKQ_cjs.GradTensor.scalar(1, { dtype: contextDtype })).reshape([batchSize, seqLenQ, this.embedDim]);
+    const contextDtype = chunkZ6BGACIH_cjs.ensureNumericDType(context.dtype, "MultiheadAttention");
+    const contextReshaped = context.transpose([0, 2, 1, 3]).mul(chunk724CXHFH_cjs.GradTensor.scalar(1, { dtype: contextDtype })).reshape([batchSize, seqLenQ, this.embedDim]);
     let output = contextReshaped.matmul(this.wO.transpose());
     if (this.bO) output = output.add(this.bO);
     if (query.ndim === 2) {
@@ -1503,33 +1514,58 @@ var TransformerEncoderLayer = class extends Module {
   dropout1;
   dropout2;
   dropout3;
-  constructor(dModel, nHead, dFF, options = {}) {
+  constructor(dModelOrOpts, nHead, dFFOrOptions, options = {}) {
     super();
+    let resolvedDModel;
+    let resolvedNHead;
+    let resolvedDFF;
+    let resolvedDropout;
+    let resolvedEps;
+    if (typeof dModelOrOpts === "object") {
+      resolvedDModel = dModelOrOpts.dModel;
+      resolvedNHead = dModelOrOpts.nHead;
+      resolvedDFF = dModelOrOpts.dFF ?? dModelOrOpts.dimFeedforward ?? 2048;
+      resolvedDropout = dModelOrOpts.dropout;
+      resolvedEps = dModelOrOpts.eps;
+    } else if (typeof dFFOrOptions === "object") {
+      resolvedDModel = dModelOrOpts;
+      resolvedNHead = nHead ?? 1;
+      resolvedDFF = dFFOrOptions.dFF ?? dFFOrOptions.dimFeedforward ?? 2048;
+      resolvedDropout = dFFOrOptions.dropout;
+      resolvedEps = dFFOrOptions.eps;
+    } else {
+      resolvedDModel = dModelOrOpts;
+      resolvedNHead = nHead ?? 1;
+      resolvedDFF = dFFOrOptions ?? 2048;
+      resolvedDropout = options.dropout;
+      resolvedEps = options.eps;
+    }
+    const dModel = resolvedDModel;
     if (!Number.isInteger(dModel) || dModel <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("dModel must be a positive integer", "dModel", dModel);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("dModel must be a positive integer", "dModel", dModel);
     }
-    if (!Number.isInteger(nHead) || nHead <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("nHead must be a positive integer", "nHead", nHead);
+    if (!Number.isInteger(resolvedNHead) || resolvedNHead <= 0) {
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("nHead must be a positive integer", "nHead", resolvedNHead);
     }
-    if (dModel % nHead !== 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
-        `dModel (${dModel}) must be divisible by nHead (${nHead})`,
+    if (dModel % resolvedNHead !== 0) {
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
+        `dModel (${dModel}) must be divisible by nHead (${resolvedNHead})`,
         "dModel",
         dModel
       );
     }
-    if (!Number.isInteger(dFF) || dFF <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("dFF must be a positive integer", "dFF", dFF);
+    if (!Number.isInteger(resolvedDFF) || resolvedDFF <= 0) {
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("dFF must be a positive integer", "dFF", resolvedDFF);
     }
-    const dropout2 = options.dropout ?? 0.1;
-    const eps = options.eps ?? 1e-5;
+    const dropout2 = resolvedDropout ?? 0.1;
+    const eps = resolvedEps ?? 1e-5;
     this.dModel = dModel;
-    this.nHead = nHead;
-    this.dFF = dFF;
+    this.nHead = resolvedNHead;
+    this.dFF = resolvedDFF;
     this.dropout = dropout2;
-    this.selfAttn = new MultiheadAttention(dModel, nHead, { dropout: dropout2 });
-    this.linear1 = new Linear(dModel, dFF);
-    this.linear2 = new Linear(dFF, dModel);
+    this.selfAttn = new MultiheadAttention(dModel, resolvedNHead, { dropout: dropout2 });
+    this.linear1 = new Linear(dModel, resolvedDFF);
+    this.linear2 = new Linear(resolvedDFF, dModel);
     this.norm1 = new LayerNorm(dModel, { eps });
     this.norm2 = new LayerNorm(dModel, { eps });
     this.dropout1 = new Dropout(dropout2);
@@ -1551,9 +1587,9 @@ var TransformerEncoderLayer = class extends Module {
    * @returns Output of same shape as input
    */
   forward(src) {
-    const input = src instanceof chunk6AE5FKKQ_cjs.GradTensor ? src : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(src);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(src) ? src : chunk724CXHFH_cjs.GradTensor.fromTensor(src);
     if (input.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("TransformerEncoderLayer does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("TransformerEncoderLayer does not support string dtype");
     }
     let src2 = this.selfAttn.forward(input, input, input);
     src2 = this.dropout1.forward(src2);
@@ -1579,7 +1615,7 @@ function normalizePair(name, value, allowZero, description) {
   const first = arr[0];
   const second = arr[1];
   if (arr.length !== 2 || first === void 0 || second === void 0 || !Number.isInteger(first) || !Number.isInteger(second) || (allowZero ? first < 0 || second < 0 : first <= 0 || second <= 0)) {
-    throw new chunkJSCDE774_cjs.InvalidParameterError(`${name} must be ${description}`, name, value);
+    throw new chunkZ6BGACIH_cjs.InvalidParameterError(`${name} must be ${description}`, name, value);
   }
   return [first, second];
 }
@@ -1595,21 +1631,21 @@ var Conv1d = class extends Module {
   constructor(inChannels, outChannels, kernelSize, options = {}) {
     super();
     if (inChannels <= 0 || !Number.isInteger(inChannels)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "inChannels must be a positive integer",
         "inChannels",
         inChannels
       );
     }
     if (outChannels <= 0 || !Number.isInteger(outChannels)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "outChannels must be a positive integer",
         "outChannels",
         outChannels
       );
     }
     if (kernelSize <= 0 || !Number.isInteger(kernelSize)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "kernelSize must be a positive integer",
         "kernelSize",
         kernelSize
@@ -1617,11 +1653,11 @@ var Conv1d = class extends Module {
     }
     const stride = options.stride ?? 1;
     if (stride <= 0 || !Number.isInteger(stride)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("stride must be a positive integer", "stride", stride);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("stride must be a positive integer", "stride", stride);
     }
     const padding = options.padding ?? 0;
     if (padding < 0 || !Number.isInteger(padding)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("padding must be a non-negative integer", "padding", padding);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("padding must be a non-negative integer", "padding", padding);
     }
     this.inChannels = inChannels;
     this.outChannels = outChannels;
@@ -1633,36 +1669,36 @@ var Conv1d = class extends Module {
   }
   initializeParameters() {
     const k = 1 / Math.sqrt(this.inChannels * this.kernelSize);
-    const weight = chunk6AE5FKKQ_cjs.randn([this.outChannels, this.inChannels, this.kernelSize]);
-    this.weight_ = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(weight, k));
+    const weight = chunk724CXHFH_cjs.randn([this.outChannels, this.inChannels, this.kernelSize]);
+    this.weight_ = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(weight, k));
     this.registerParameter("weight", this.weight_);
     if (this.bias) {
-      const biasInit = chunk6AE5FKKQ_cjs.randn([this.outChannels]);
-      this.bias_ = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(biasInit, k));
+      const biasInit = chunk724CXHFH_cjs.randn([this.outChannels]);
+      this.bias_ = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(biasInit, k));
       this.registerParameter("bias", this.bias_);
     }
   }
   forward(x) {
-    const input = x instanceof chunk6AE5FKKQ_cjs.GradTensor ? x : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(x);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(x) ? x : chunk724CXHFH_cjs.GradTensor.fromTensor(x);
     if (input.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("String tensors are not supported");
+      throw new chunkZ6BGACIH_cjs.DTypeError("String tensors are not supported");
     }
     if (input.ndim !== 3) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Conv1d expects 3D input (batch, channels, length), got ${input.ndim}D`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Conv1d expects 3D input (batch, channels, length), got ${input.ndim}D`);
     }
     const batch = input.shape[0] ?? 0;
     const inC = input.shape[1] ?? 0;
     const inL = input.shape[2] ?? 0;
     if (inC !== this.inChannels) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected ${this.inChannels} input channels, got ${inC}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected ${this.inChannels} input channels, got ${inC}`);
     }
     const weight = this.weight_;
-    if (!weight) throw new chunkJSCDE774_cjs.NotFittedError("Weight not initialized");
+    if (!weight) throw new chunkZ6BGACIH_cjs.NotFittedError("Weight not initialized");
     const input2d = input.reshape([batch, inC, 1, inL]);
     const kernelSize = [1, this.kernelSize];
     const stride = [1, this.stride];
     const padding = [0, this.padding];
-    const cols = chunk6AE5FKKQ_cjs.im2col2(input2d, kernelSize, stride, padding);
+    const cols = chunk724CXHFH_cjs.im2col2(input2d, kernelSize, stride, padding);
     const weightFlat = weight.reshape([this.outChannels, this.inChannels * this.kernelSize]);
     const out = cols.matmul(weightFlat.transpose());
     const outTransposed = out.transpose([0, 2, 1]);
@@ -1674,7 +1710,7 @@ var Conv1d = class extends Module {
   }
   get weight() {
     if (!this.weight_) {
-      throw new chunkJSCDE774_cjs.NotFittedError("Weight not initialized");
+      throw new chunkZ6BGACIH_cjs.NotFittedError("Weight not initialized");
     }
     return this.weight_;
   }
@@ -1691,14 +1727,14 @@ var Conv2d = class extends Module {
   constructor(inChannels, outChannels, kernelSize, options = {}) {
     super();
     if (inChannels <= 0 || !Number.isInteger(inChannels)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "inChannels must be a positive integer",
         "inChannels",
         inChannels
       );
     }
     if (outChannels <= 0 || !Number.isInteger(outChannels)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "outChannels must be a positive integer",
         "outChannels",
         outChannels
@@ -1736,22 +1772,22 @@ var Conv2d = class extends Module {
     const kH = this.kernelSize[0] ?? 1;
     const kW = this.kernelSize[1] ?? 1;
     const k = 1 / Math.sqrt(this.inChannels * kH * kW);
-    const weight = chunk6AE5FKKQ_cjs.randn([this.outChannels, this.inChannels, kH, kW]);
-    this.weight_ = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(weight, k));
+    const weight = chunk724CXHFH_cjs.randn([this.outChannels, this.inChannels, kH, kW]);
+    this.weight_ = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(weight, k));
     this.registerParameter("weight", this.weight_);
     if (this.useBias) {
-      const biasInit = chunk6AE5FKKQ_cjs.randn([this.outChannels]);
-      this.bias_ = chunk6AE5FKKQ_cjs.parameter(chunk6AE5FKKQ_cjs.mulScalar(biasInit, k));
+      const biasInit = chunk724CXHFH_cjs.randn([this.outChannels]);
+      this.bias_ = chunk724CXHFH_cjs.parameter(chunk724CXHFH_cjs.mulScalar(biasInit, k));
       this.registerParameter("bias", this.bias_);
     }
   }
   forward(x) {
-    const input = x instanceof chunk6AE5FKKQ_cjs.GradTensor ? x : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(x);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(x) ? x : chunk724CXHFH_cjs.GradTensor.fromTensor(x);
     if (input.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("String tensors are not supported");
+      throw new chunkZ6BGACIH_cjs.DTypeError("String tensors are not supported");
     }
     if (input.ndim !== 4) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `Conv2d expects 4D input (batch, channels, height, width), got ${input.ndim}D`
       );
     }
@@ -1760,14 +1796,14 @@ var Conv2d = class extends Module {
     const inH = input.shape[2] ?? 0;
     const inW = input.shape[3] ?? 0;
     if (inC !== this.inChannels) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected ${this.inChannels} input channels, got ${inC}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected ${this.inChannels} input channels, got ${inC}`);
     }
     const weight = this.weight_;
-    if (!weight) throw new chunkJSCDE774_cjs.NotFittedError("Weight not initialized");
+    if (!weight) throw new chunkZ6BGACIH_cjs.NotFittedError("Weight not initialized");
     const [kH, kW] = this.kernelSize;
     const [sH, sW] = this.stride;
     const [pH, pW] = this.padding;
-    const cols = chunk6AE5FKKQ_cjs.im2col2(input, [kH, kW], [sH, sW], [pH, pW]);
+    const cols = chunk724CXHFH_cjs.im2col2(input, [kH, kW], [sH, sW], [pH, pW]);
     const outH = Math.floor((inH + 2 * pH - kH) / sH) + 1;
     const outW = Math.floor((inW + 2 * pW - kW) / sW) + 1;
     const weightFlat = weight.reshape([this.outChannels, this.inChannels * kH * kW]);
@@ -1782,7 +1818,7 @@ var Conv2d = class extends Module {
   }
   get weight() {
     if (!this.weight_) {
-      throw new chunkJSCDE774_cjs.NotFittedError("Weight not initialized");
+      throw new chunkZ6BGACIH_cjs.NotFittedError("Weight not initialized");
     }
     return this.weight_;
   }
@@ -1816,12 +1852,12 @@ var MaxPool2d = class extends Module {
     this.padding = paddingArr;
   }
   forward(x) {
-    const input = x instanceof chunk6AE5FKKQ_cjs.GradTensor ? x : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(x);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(x) ? x : chunk724CXHFH_cjs.GradTensor.fromTensor(x);
     if (input.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("String tensors are not supported");
+      throw new chunkZ6BGACIH_cjs.DTypeError("String tensors are not supported");
     }
     if (input.ndim !== 4) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `MaxPool2d expects 4D input (batch, channels, height, width), got ${input.ndim}D`
       );
     }
@@ -1833,7 +1869,7 @@ var MaxPool2d = class extends Module {
     const [sH, sW] = this.stride;
     const [pH, pW] = this.padding;
     const inputReshaped = input.reshape([batch * channels, 1, inH, inW]);
-    const cols = chunk6AE5FKKQ_cjs.im2col2(inputReshaped, [kH, kW], [sH, sW], [pH, pW]);
+    const cols = chunk724CXHFH_cjs.im2col2(inputReshaped, [kH, kW], [sH, sW], [pH, pW]);
     const maxVals = cols.max(2);
     const outH = Math.floor((inH + 2 * pH - kH) / sH) + 1;
     const outW = Math.floor((inW + 2 * pW - kW) / sW) + 1;
@@ -1869,12 +1905,12 @@ var AvgPool2d = class extends Module {
     this.padding = paddingArr;
   }
   forward(x) {
-    const input = x instanceof chunk6AE5FKKQ_cjs.GradTensor ? x : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(x);
+    const input = chunk724CXHFH_cjs.GradTensor.isGradTensor(x) ? x : chunk724CXHFH_cjs.GradTensor.fromTensor(x);
     if (input.dtype === "string") {
-      throw new chunkJSCDE774_cjs.DTypeError("String tensors are not supported");
+      throw new chunkZ6BGACIH_cjs.DTypeError("String tensors are not supported");
     }
     if (input.ndim !== 4) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `AvgPool2d expects 4D input (batch, channels, height, width), got ${input.ndim}D`
       );
     }
@@ -1886,7 +1922,7 @@ var AvgPool2d = class extends Module {
     const [sH, sW] = this.stride;
     const [pH, pW] = this.padding;
     const inputReshaped = input.reshape([batch * channels, 1, inH, inW]);
-    const cols = chunk6AE5FKKQ_cjs.im2col2(inputReshaped, [kH, kW], [sH, sW], [pH, pW]);
+    const cols = chunk724CXHFH_cjs.im2col2(inputReshaped, [kH, kW], [sH, sW], [pH, pW]);
     const meanVals = cols.mean(2);
     const outH = Math.floor((inH + 2 * pH - kH) / sH) + 1;
     const outW = Math.floor((inW + 2 * pW - kW) / sW) + 1;
@@ -1897,25 +1933,25 @@ var AvgPool2d = class extends Module {
 // src/nn/layers/recurrent.ts
 function ensureFloatTensor(t, context) {
   if (t.dtype === "string") {
-    throw new chunkJSCDE774_cjs.DTypeError(`${context} does not support string dtype`);
+    throw new chunkZ6BGACIH_cjs.DTypeError(`${context} does not support string dtype`);
   }
   if (t.dtype !== "float32" && t.dtype !== "float64") {
-    throw new chunkJSCDE774_cjs.DTypeError(`${context} expects float32 or float64 dtype`);
+    throw new chunkZ6BGACIH_cjs.DTypeError(`${context} expects float32 or float64 dtype`);
   }
 }
 function readNumeric(t, offset) {
   const data = t.data;
   if (Array.isArray(data)) {
-    throw new chunkJSCDE774_cjs.DTypeError("String tensors are not supported");
+    throw new chunkZ6BGACIH_cjs.DTypeError("String tensors are not supported");
   }
-  return chunkJSCDE774_cjs.getElementAsNumber(data, offset);
+  return chunkZ6BGACIH_cjs.getElementAsNumber(data, offset);
 }
 function createFloatBuffer(size, dtype) {
   return dtype === "float64" ? new Float64Array(size) : new Float32Array(size);
 }
 function validatePositiveInt(name, value) {
   if (!Number.isInteger(value) || value <= 0) {
-    throw new chunkJSCDE774_cjs.InvalidParameterError(`${name} must be a positive integer`, name, value);
+    throw new chunkZ6BGACIH_cjs.InvalidParameterError(`${name} must be a positive integer`, name, value);
   }
 }
 function parseInput(input, batchFirst) {
@@ -1933,7 +1969,7 @@ function parseInput(input, batchFirst) {
     };
   }
   if (input.ndim !== 3) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Recurrent layers expect 2D or 3D input; got ndim=${input.ndim}`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Recurrent layers expect 2D or 3D input; got ndim=${input.ndim}`);
   }
   if (batchFirst) {
     return {
@@ -1966,7 +2002,7 @@ function outputIndex(batchFirst, isUnbatched, batch, seqLen, hiddenSize, b, t, j
   return t * (batch * hiddenSize) + b * hiddenSize + j;
 }
 function extractTensor(arg, _name) {
-  if (arg instanceof chunk6AE5FKKQ_cjs.GradTensor) {
+  if (chunk724CXHFH_cjs.GradTensor.isGradTensor(arg)) {
     return arg.tensor;
   }
   return arg;
@@ -1982,10 +2018,10 @@ function buildState(state, numLayers, batch, hiddenSize, isUnbatched, name) {
   ensureFloatTensor(state, name);
   if (state.ndim === 2) {
     if (!isUnbatched) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected ${name} with 3 dimensions for batched input`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected ${name} with 3 dimensions for batched input`);
     }
     if ((state.shape[0] ?? 0) !== numLayers || (state.shape[1] ?? 0) !== hiddenSize) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `Expected ${name} shape [${numLayers}, ${hiddenSize}], got [${state.shape.join(", ")}]`
       );
     }
@@ -1994,7 +2030,7 @@ function buildState(state, numLayers, batch, hiddenSize, isUnbatched, name) {
     for (let l = 0; l < numLayers; l++) {
       const layerState = result[l];
       if (!layerState) {
-        throw new chunkJSCDE774_cjs.ShapeError(`Internal error: missing ${name} layer state`);
+        throw new chunkZ6BGACIH_cjs.ShapeError(`Internal error: missing ${name} layer state`);
       }
       const base = state.offset + l * stride02;
       for (let j = 0; j < hiddenSize; j++) {
@@ -2004,12 +2040,12 @@ function buildState(state, numLayers, batch, hiddenSize, isUnbatched, name) {
     return result;
   }
   if (state.ndim !== 3) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Expected ${name} with 2 or 3 dimensions; got ndim=${state.ndim}`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Expected ${name} with 2 or 3 dimensions; got ndim=${state.ndim}`);
   }
   const expectedBatch = isUnbatched ? 1 : batch;
   if ((state.shape[0] ?? 0) !== numLayers || (state.shape[1] ?? 0) !== expectedBatch || (state.shape[2] ?? 0) !== hiddenSize) {
     const expected = isUnbatched ? [numLayers, 1, hiddenSize] : [numLayers, batch, hiddenSize];
-    throw new chunkJSCDE774_cjs.ShapeError(
+    throw new chunkZ6BGACIH_cjs.ShapeError(
       `Expected ${name} shape [${expected.join(", ")}], got [${state.shape.join(", ")}]`
     );
   }
@@ -2019,7 +2055,7 @@ function buildState(state, numLayers, batch, hiddenSize, isUnbatched, name) {
   for (let l = 0; l < numLayers; l++) {
     const layerState = result[l];
     if (!layerState) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Internal error: missing ${name} layer state`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Internal error: missing ${name} layer state`);
     }
     const baseLayer = state.offset + l * stride0;
     for (let b = 0; b < batch; b++) {
@@ -2038,13 +2074,13 @@ function packState(state, numLayers, batch, hiddenSize, dtype, device, isUnbatch
     for (let l = 0; l < numLayers; l++) {
       const layer = state[l];
       if (!layer) {
-        throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing packed state layer");
+        throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing packed state layer");
       }
       for (let j = 0; j < hiddenSize; j++) {
         data[l * hiddenSize + j] = layer[j] ?? 0;
       }
     }
-    return chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+    return chunk724CXHFH_cjs.Tensor.fromTypedArray({
       data,
       shape: [numLayers, hiddenSize],
       dtype,
@@ -2054,7 +2090,7 @@ function packState(state, numLayers, batch, hiddenSize, dtype, device, isUnbatch
   for (let l = 0; l < numLayers; l++) {
     const layer = state[l];
     if (!layer) {
-      throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing packed state layer");
+      throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing packed state layer");
     }
     const layerOffset = l * batch * hiddenSize;
     for (let b = 0; b < batch; b++) {
@@ -2064,7 +2100,7 @@ function packState(state, numLayers, batch, hiddenSize, dtype, device, isUnbatch
       }
     }
   }
-  return chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+  return chunk724CXHFH_cjs.Tensor.fromTypedArray({
     data,
     shape: [numLayers, batch, hiddenSize],
     dtype,
@@ -2101,19 +2137,19 @@ var RNN = class extends Module {
     this.biasHh = [];
     for (let layer = 0; layer < this.numLayers; layer++) {
       const inputDim = layer === 0 ? inputSize : hiddenSize;
-      const wIh = chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([hiddenSize, inputDim]), stdv);
-      const wHh = chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([hiddenSize, hiddenSize]), stdv);
+      const wIh = chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([hiddenSize, inputDim]), stdv);
+      const wHh = chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([hiddenSize, hiddenSize]), stdv);
       this.weightsIh.push(wIh);
       this.weightsHh.push(wHh);
-      this.registerParameter(`weight_ih_l${layer}`, chunk6AE5FKKQ_cjs.parameter(wIh));
-      this.registerParameter(`weight_hh_l${layer}`, chunk6AE5FKKQ_cjs.parameter(wHh));
+      this.registerParameter(`weight_ih_l${layer}`, chunk724CXHFH_cjs.parameter(wIh));
+      this.registerParameter(`weight_hh_l${layer}`, chunk724CXHFH_cjs.parameter(wHh));
       if (this.bias) {
-        const bIh = chunk6AE5FKKQ_cjs.zeros([hiddenSize]);
-        const bHh = chunk6AE5FKKQ_cjs.zeros([hiddenSize]);
+        const bIh = chunk724CXHFH_cjs.zeros([hiddenSize]);
+        const bHh = chunk724CXHFH_cjs.zeros([hiddenSize]);
         this.biasIh.push(bIh);
         this.biasHh.push(bHh);
-        this.registerParameter(`bias_ih_l${layer}`, chunk6AE5FKKQ_cjs.parameter(bIh));
-        this.registerParameter(`bias_hh_l${layer}`, chunk6AE5FKKQ_cjs.parameter(bHh));
+        this.registerParameter(`bias_ih_l${layer}`, chunk724CXHFH_cjs.parameter(bIh));
+        this.registerParameter(`bias_hh_l${layer}`, chunk724CXHFH_cjs.parameter(bHh));
       }
     }
   }
@@ -2125,13 +2161,13 @@ var RNN = class extends Module {
     const parsed = parseInput(input, this.batchFirst);
     const { batch, seqLen, inputDim, isUnbatched, batchStride, seqStride, featStride } = parsed;
     if (inputDim !== this.inputSize) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);
     }
     if (seqLen <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Sequence length must be positive", "seqLen", seqLen);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Sequence length must be positive", "seqLen", seqLen);
     }
     if (!isUnbatched && batch <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Batch size must be positive", "batch", batch);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Batch size must be positive", "batch", batch);
     }
     const h = buildState(hx, this.numLayers, batch, this.hiddenSize, isUnbatched, "hx");
     const outSize = (isUnbatched ? seqLen : batch * seqLen) * this.hiddenSize;
@@ -2148,13 +2184,13 @@ var RNN = class extends Module {
           const wIh = this.weightsIh[l];
           const wHh = this.weightsHh[l];
           if (!wIh || !wHh) {
-            throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing RNN weights");
+            throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing RNN weights");
           }
           const curInputSize = l === 0 ? this.inputSize : this.hiddenSize;
           const newH = new Float64Array(this.hiddenSize);
           const hLayer = h[l];
           if (!hLayer) {
-            throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing RNN hidden state");
+            throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing RNN hidden state");
           }
           const wIhStride0 = wIh.strides[0] ?? 0;
           const wIhStride1 = wIh.strides[1] ?? 0;
@@ -2202,7 +2238,7 @@ var RNN = class extends Module {
     }
     const outShape = isUnbatched ? [seqLen, this.hiddenSize] : this.batchFirst ? [batch, seqLen, this.hiddenSize] : [seqLen, batch, this.hiddenSize];
     return {
-      output: chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+      output: chunk724CXHFH_cjs.Tensor.fromTypedArray({
         data: out,
         shape: outShape,
         dtype: input.dtype,
@@ -2221,11 +2257,11 @@ var RNN = class extends Module {
   }
   forward(...inputs) {
     if (inputs.length < 1 || inputs.length > 2) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("RNN.forward expects 1 or 2 inputs", "inputs", inputs.length);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("RNN.forward expects 1 or 2 inputs", "inputs", inputs.length);
     }
     const inputArg = inputs[0];
     if (inputArg === void 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("RNN.forward requires an input tensor", "input", inputArg);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("RNN.forward requires an input tensor", "input", inputArg);
     }
     const input = extractTensor(inputArg);
     const hxArg = inputs.length === 2 ? inputs[1] : void 0;
@@ -2274,19 +2310,19 @@ var LSTM = class extends Module {
     this.biasHh = [];
     for (let layer = 0; layer < this.numLayers; layer++) {
       const inputDim = layer === 0 ? inputSize : hiddenSize;
-      const wIh = chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([4 * hiddenSize, inputDim]), stdv);
-      const wHh = chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([4 * hiddenSize, hiddenSize]), stdv);
+      const wIh = chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([4 * hiddenSize, inputDim]), stdv);
+      const wHh = chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([4 * hiddenSize, hiddenSize]), stdv);
       this.weightsIh.push(wIh);
       this.weightsHh.push(wHh);
-      this.registerParameter(`weight_ih_l${layer}`, chunk6AE5FKKQ_cjs.parameter(wIh));
-      this.registerParameter(`weight_hh_l${layer}`, chunk6AE5FKKQ_cjs.parameter(wHh));
+      this.registerParameter(`weight_ih_l${layer}`, chunk724CXHFH_cjs.parameter(wIh));
+      this.registerParameter(`weight_hh_l${layer}`, chunk724CXHFH_cjs.parameter(wHh));
       if (this.bias) {
-        const bIh = chunk6AE5FKKQ_cjs.zeros([4 * hiddenSize]);
-        const bHh = chunk6AE5FKKQ_cjs.zeros([4 * hiddenSize]);
+        const bIh = chunk724CXHFH_cjs.zeros([4 * hiddenSize]);
+        const bHh = chunk724CXHFH_cjs.zeros([4 * hiddenSize]);
         this.biasIh.push(bIh);
         this.biasHh.push(bHh);
-        this.registerParameter(`bias_ih_l${layer}`, chunk6AE5FKKQ_cjs.parameter(bIh));
-        this.registerParameter(`bias_hh_l${layer}`, chunk6AE5FKKQ_cjs.parameter(bHh));
+        this.registerParameter(`bias_ih_l${layer}`, chunk724CXHFH_cjs.parameter(bIh));
+        this.registerParameter(`bias_hh_l${layer}`, chunk724CXHFH_cjs.parameter(bHh));
       }
     }
   }
@@ -2298,13 +2334,13 @@ var LSTM = class extends Module {
     const parsed = parseInput(input, this.batchFirst);
     const { batch, seqLen, inputDim, isUnbatched, batchStride, seqStride, featStride } = parsed;
     if (inputDim !== this.inputSize) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);
     }
     if (seqLen <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Sequence length must be positive", "seqLen", seqLen);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Sequence length must be positive", "seqLen", seqLen);
     }
     if (!isUnbatched && batch <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Batch size must be positive", "batch", batch);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Batch size must be positive", "batch", batch);
     }
     const h = buildState(hx, this.numLayers, batch, this.hiddenSize, isUnbatched, "hx");
     const c = buildState(cx, this.numLayers, batch, this.hiddenSize, isUnbatched, "cx");
@@ -2323,13 +2359,13 @@ var LSTM = class extends Module {
           const wIh = this.weightsIh[l];
           const wHh = this.weightsHh[l];
           if (!wIh || !wHh) {
-            throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing LSTM weights");
+            throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing LSTM weights");
           }
           const curInputSize = l === 0 ? this.inputSize : this.hiddenSize;
           const hLayer = h[l];
           const cLayer = c[l];
           if (!hLayer || !cLayer) {
-            throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing LSTM state");
+            throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing LSTM state");
           }
           const wIhStride0 = wIh.strides[0] ?? 0;
           const wIhStride1 = wIh.strides[1] ?? 0;
@@ -2391,7 +2427,7 @@ var LSTM = class extends Module {
     }
     const outShape = isUnbatched ? [seqLen, this.hiddenSize] : this.batchFirst ? [batch, seqLen, this.hiddenSize] : [seqLen, batch, this.hiddenSize];
     return {
-      output: chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+      output: chunk724CXHFH_cjs.Tensor.fromTypedArray({
         data: out,
         shape: outShape,
         dtype: input.dtype,
@@ -2419,7 +2455,7 @@ var LSTM = class extends Module {
   }
   forward(...inputs) {
     if (inputs.length < 1 || inputs.length > 3) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         "LSTM.forward expects 1 to 3 inputs",
         "inputs",
         inputs.length
@@ -2427,7 +2463,7 @@ var LSTM = class extends Module {
     }
     const inputArg = inputs[0];
     if (inputArg === void 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("LSTM.forward requires an input tensor", "input", inputArg);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("LSTM.forward requires an input tensor", "input", inputArg);
     }
     const input = extractTensor(inputArg);
     const hxArg = inputs.length >= 2 ? inputs[1] : void 0;
@@ -2479,19 +2515,19 @@ var GRU = class extends Module {
     this.biasHh = [];
     for (let layer = 0; layer < this.numLayers; layer++) {
       const inputDim = layer === 0 ? inputSize : hiddenSize;
-      const wIh = chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([3 * hiddenSize, inputDim]), stdv);
-      const wHh = chunk6AE5FKKQ_cjs.mulScalar(chunk6AE5FKKQ_cjs.randn([3 * hiddenSize, hiddenSize]), stdv);
+      const wIh = chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([3 * hiddenSize, inputDim]), stdv);
+      const wHh = chunk724CXHFH_cjs.mulScalar(chunk724CXHFH_cjs.randn([3 * hiddenSize, hiddenSize]), stdv);
       this.weightsIh.push(wIh);
       this.weightsHh.push(wHh);
-      this.registerParameter(`weight_ih_l${layer}`, chunk6AE5FKKQ_cjs.parameter(wIh));
-      this.registerParameter(`weight_hh_l${layer}`, chunk6AE5FKKQ_cjs.parameter(wHh));
+      this.registerParameter(`weight_ih_l${layer}`, chunk724CXHFH_cjs.parameter(wIh));
+      this.registerParameter(`weight_hh_l${layer}`, chunk724CXHFH_cjs.parameter(wHh));
       if (this.bias) {
-        const bIh = chunk6AE5FKKQ_cjs.zeros([3 * hiddenSize]);
-        const bHh = chunk6AE5FKKQ_cjs.zeros([3 * hiddenSize]);
+        const bIh = chunk724CXHFH_cjs.zeros([3 * hiddenSize]);
+        const bHh = chunk724CXHFH_cjs.zeros([3 * hiddenSize]);
         this.biasIh.push(bIh);
         this.biasHh.push(bHh);
-        this.registerParameter(`bias_ih_l${layer}`, chunk6AE5FKKQ_cjs.parameter(bIh));
-        this.registerParameter(`bias_hh_l${layer}`, chunk6AE5FKKQ_cjs.parameter(bHh));
+        this.registerParameter(`bias_ih_l${layer}`, chunk724CXHFH_cjs.parameter(bIh));
+        this.registerParameter(`bias_hh_l${layer}`, chunk724CXHFH_cjs.parameter(bHh));
       }
     }
   }
@@ -2503,13 +2539,13 @@ var GRU = class extends Module {
     const parsed = parseInput(input, this.batchFirst);
     const { batch, seqLen, inputDim, isUnbatched, batchStride, seqStride, featStride } = parsed;
     if (inputDim !== this.inputSize) {
-      throw new chunkJSCDE774_cjs.ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);
+      throw new chunkZ6BGACIH_cjs.ShapeError(`Expected input size ${this.inputSize}, got ${inputDim}`);
     }
     if (seqLen <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Sequence length must be positive", "seqLen", seqLen);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Sequence length must be positive", "seqLen", seqLen);
     }
     if (!isUnbatched && batch <= 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("Batch size must be positive", "batch", batch);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("Batch size must be positive", "batch", batch);
     }
     const h = buildState(hx, this.numLayers, batch, this.hiddenSize, isUnbatched, "hx");
     const outSize = (isUnbatched ? seqLen : batch * seqLen) * this.hiddenSize;
@@ -2528,12 +2564,12 @@ var GRU = class extends Module {
           const wIh = this.weightsIh[l];
           const wHh = this.weightsHh[l];
           if (!wIh || !wHh) {
-            throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing GRU weights");
+            throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing GRU weights");
           }
           const curInputSize = l === 0 ? this.inputSize : this.hiddenSize;
           const hLayer = h[l];
           if (!hLayer) {
-            throw new chunkJSCDE774_cjs.ShapeError("Internal error: missing GRU hidden state");
+            throw new chunkZ6BGACIH_cjs.ShapeError("Internal error: missing GRU hidden state");
           }
           const wIhStride0 = wIh.strides[0] ?? 0;
           const wIhStride1 = wIh.strides[1] ?? 0;
@@ -2594,7 +2630,7 @@ var GRU = class extends Module {
     }
     const outShape = isUnbatched ? [seqLen, this.hiddenSize] : this.batchFirst ? [batch, seqLen, this.hiddenSize] : [seqLen, batch, this.hiddenSize];
     return {
-      output: chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+      output: chunk724CXHFH_cjs.Tensor.fromTypedArray({
         data: out,
         shape: outShape,
         dtype: input.dtype,
@@ -2613,11 +2649,11 @@ var GRU = class extends Module {
   }
   forward(...inputs) {
     if (inputs.length < 1 || inputs.length > 2) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("GRU.forward expects 1 or 2 inputs", "inputs", inputs.length);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("GRU.forward expects 1 or 2 inputs", "inputs", inputs.length);
     }
     const inputArg = inputs[0];
     if (inputArg === void 0) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError("GRU.forward requires an input tensor", "input", inputArg);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError("GRU.forward requires an input tensor", "input", inputArg);
     }
     const input = extractTensor(inputArg);
     const hxArg = inputs.length === 2 ? inputs[1] : void 0;
@@ -2645,7 +2681,7 @@ function toOneHot(indices, numClasses) {
   const outData = new Float32Array(nSamples * numClasses);
   const data = indices.data;
   if (Array.isArray(data)) {
-    throw new chunkJSCDE774_cjs.DTypeError("crossEntropyLoss target indices must be numeric");
+    throw new chunkZ6BGACIH_cjs.DTypeError("crossEntropyLoss target indices must be numeric");
   }
   const stride0 = indices.strides[0] ?? 0;
   const base = indices.offset;
@@ -2653,10 +2689,10 @@ function toOneHot(indices, numClasses) {
     const offset = base + i * stride0;
     let idx;
     if (data instanceof BigInt64Array) {
-      const raw = chunkJSCDE774_cjs.getBigIntElement(data, offset);
+      const raw = chunkZ6BGACIH_cjs.getBigIntElement(data, offset);
       const asNumber = Number(raw);
       if (!Number.isSafeInteger(asNumber)) {
-        throw new chunkJSCDE774_cjs.InvalidParameterError(
+        throw new chunkZ6BGACIH_cjs.InvalidParameterError(
           `Class index ${raw.toString()} exceeds safe integer range`,
           "target",
           raw.toString()
@@ -2664,13 +2700,13 @@ function toOneHot(indices, numClasses) {
       }
       idx = asNumber;
     } else {
-      idx = Number(chunkJSCDE774_cjs.getNumericElement(data, offset));
+      idx = Number(chunkZ6BGACIH_cjs.getNumericElement(data, offset));
     }
     if (!Number.isFinite(idx) || !Number.isInteger(idx)) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(`Class index ${idx} is not a valid integer`, "target", idx);
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(`Class index ${idx} is not a valid integer`, "target", idx);
     }
     if (idx < 0 || idx >= numClasses) {
-      throw new chunkJSCDE774_cjs.InvalidParameterError(
+      throw new chunkZ6BGACIH_cjs.InvalidParameterError(
         `Class index ${idx} out of range [0, ${numClasses})`,
         "target",
         idx
@@ -2678,7 +2714,7 @@ function toOneHot(indices, numClasses) {
     }
     outData[i * numClasses + idx] = 1;
   }
-  return chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+  return chunk724CXHFH_cjs.Tensor.fromTypedArray({
     data: outData,
     shape: [nSamples, numClasses],
     dtype: "float32",
@@ -2686,62 +2722,59 @@ function toOneHot(indices, numClasses) {
   });
 }
 function crossEntropyLoss(input, target) {
-  const yPred = input instanceof chunk6AE5FKKQ_cjs.GradTensor ? input : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(input);
-  const targetIsGrad = target instanceof chunk6AE5FKKQ_cjs.GradTensor;
-  const yTrue = target instanceof chunk6AE5FKKQ_cjs.GradTensor ? target : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(target, { requiresGrad: false });
+  const yPred = chunk724CXHFH_cjs.GradTensor.isGradTensor(input) ? input : chunk724CXHFH_cjs.GradTensor.fromTensor(input);
+  const targetIsGrad = chunk724CXHFH_cjs.GradTensor.isGradTensor(target);
+  const yTrue = chunk724CXHFH_cjs.GradTensor.isGradTensor(target) ? target : chunk724CXHFH_cjs.GradTensor.fromTensor(target, { requiresGrad: false });
   if (yPred.ndim !== 2) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Input must be 2-dimensional (batch, classes); got ${yPred.ndim}`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Input must be 2-dimensional (batch, classes); got ${yPred.ndim}`);
   }
   const nSamples = yPred.shape[0] ?? 0;
   const nClasses = yPred.shape[1] ?? 0;
   let targetTensor = yTrue;
   if (yTrue.ndim === 1) {
-    if (targetIsGrad) {
-      throw new chunkJSCDE774_cjs.ShapeError("Target must be 2-dimensional when provided as GradTensor");
-    }
     if (yTrue.shape[0] !== nSamples) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         `Target must have same number of samples as input; got ${yTrue.shape[0]} and ${nSamples}`
       );
     }
     const oneHot = toOneHot(yTrue.tensor, nClasses);
-    targetTensor = chunk6AE5FKKQ_cjs.GradTensor.fromTensor(oneHot, { requiresGrad: false });
+    targetTensor = chunk724CXHFH_cjs.GradTensor.fromTensor(oneHot, { requiresGrad: false });
   } else if (yTrue.ndim === 2) {
     if (yTrue.shape[0] !== nSamples || yTrue.shape[1] !== nClasses) {
-      throw new chunkJSCDE774_cjs.ShapeError(
+      throw new chunkZ6BGACIH_cjs.ShapeError(
         "Target must be 1-dimensional class indices or have the same shape as input"
       );
     }
   } else {
-    throw new chunkJSCDE774_cjs.ShapeError(`Target must be 1D (indices) or 2D (probs); got ${yTrue.ndim}D`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Target must be 1D (indices) or 2D (probs); got ${yTrue.ndim}D`);
   }
-  const logProbs = chunk6AE5FKKQ_cjs.logSoftmax2(yPred, 1);
+  const logProbs = chunk724CXHFH_cjs.logSoftmax2(yPred, 1);
   const weighted = logProbs.mul(targetTensor);
   const sampleLoss = weighted.sum(1);
   const meanLoss = sampleLoss.mean().neg();
-  if (!(input instanceof chunk6AE5FKKQ_cjs.GradTensor) && !targetIsGrad) {
+  if (!chunk724CXHFH_cjs.GradTensor.isGradTensor(input) && !targetIsGrad) {
     const data = meanLoss.tensor.data;
     if (Array.isArray(data)) {
-      throw new chunkJSCDE774_cjs.DTypeError("crossEntropyLoss does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("crossEntropyLoss does not support string dtype");
     }
     if (data instanceof BigInt64Array) {
-      const raw = chunkJSCDE774_cjs.getBigIntElement(data, meanLoss.tensor.offset);
+      const raw = chunkZ6BGACIH_cjs.getBigIntElement(data, meanLoss.tensor.offset);
       return Number(raw);
     }
-    return chunkJSCDE774_cjs.getNumericElement(data, meanLoss.tensor.offset);
+    return chunkZ6BGACIH_cjs.getNumericElement(data, meanLoss.tensor.offset);
   }
   return meanLoss;
 }
 function binaryCrossEntropyWithLogitsLoss(input, target) {
-  const yPred = input instanceof chunk6AE5FKKQ_cjs.GradTensor ? input : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(input);
-  const yTrue = target instanceof chunk6AE5FKKQ_cjs.GradTensor ? target : chunk6AE5FKKQ_cjs.GradTensor.fromTensor(target, { requiresGrad: false });
+  const yPred = chunk724CXHFH_cjs.GradTensor.isGradTensor(input) ? input : chunk724CXHFH_cjs.GradTensor.fromTensor(input);
+  const yTrue = chunk724CXHFH_cjs.GradTensor.isGradTensor(target) ? target : chunk724CXHFH_cjs.GradTensor.fromTensor(target, { requiresGrad: false });
   let pred = yPred;
   let truth = yTrue;
   if (pred.ndim !== 1 && pred.ndim !== 2) {
-    throw new chunkJSCDE774_cjs.ShapeError("Input must be 1 or 2-dimensional");
+    throw new chunkZ6BGACIH_cjs.ShapeError("Input must be 1 or 2-dimensional");
   }
   if (truth.ndim !== 1 && truth.ndim !== 2) {
-    throw new chunkJSCDE774_cjs.ShapeError("Target must be 1 or 2-dimensional");
+    throw new chunkZ6BGACIH_cjs.ShapeError("Target must be 1 or 2-dimensional");
   }
   if (pred.ndim === 1) {
     pred = pred.reshape([pred.shape[0] ?? 0, 1]);
@@ -2750,17 +2783,17 @@ function binaryCrossEntropyWithLogitsLoss(input, target) {
     truth = truth.reshape([truth.shape[0] ?? 0, 1]);
   }
   if (pred.ndim !== 2 || pred.shape[1] !== 1) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Input must have shape (N,) or (N, 1)`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Input must have shape (N,) or (N, 1)`);
   }
   if (truth.ndim !== 2 || truth.shape[1] !== 1) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Target must be 1-dimensional or have shape (N, 1)`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Target must be 1-dimensional or have shape (N, 1)`);
   }
   if ((pred.shape[0] ?? 0) !== (truth.shape[0] ?? 0)) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Batch size mismatch`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Batch size mismatch`);
   }
   const predDtype = pred.dtype;
   if (predDtype === "string") {
-    throw new chunkJSCDE774_cjs.DTypeError("Binary cross entropy does not support string dtype");
+    throw new chunkZ6BGACIH_cjs.DTypeError("Binary cross entropy does not support string dtype");
   }
   const term1 = pred.relu();
   const term2 = pred.mul(truth);
@@ -2769,21 +2802,21 @@ function binaryCrossEntropyWithLogitsLoss(input, target) {
   const expNegAbs = absPred.neg().exp();
   const scalarDtype = expNegAbs.dtype;
   if (scalarDtype === "string") {
-    throw new chunkJSCDE774_cjs.DTypeError("binaryCrossEntropyWithLogitsLoss does not support string dtype");
+    throw new chunkZ6BGACIH_cjs.DTypeError("binaryCrossEntropyWithLogitsLoss does not support string dtype");
   }
-  const one = chunk6AE5FKKQ_cjs.GradTensor.scalar(1, { dtype: scalarDtype });
+  const one = chunk724CXHFH_cjs.GradTensor.scalar(1, { dtype: scalarDtype });
   const term3 = one.add(expNegAbs).log();
   const loss = term1.sub(term2).add(term3).mean();
-  if (!(input instanceof chunk6AE5FKKQ_cjs.GradTensor) && !(target instanceof chunk6AE5FKKQ_cjs.GradTensor)) {
+  if (!chunk724CXHFH_cjs.GradTensor.isGradTensor(input) && !chunk724CXHFH_cjs.GradTensor.isGradTensor(target)) {
     const data = loss.tensor.data;
     if (Array.isArray(data)) {
-      throw new chunkJSCDE774_cjs.DTypeError("binaryCrossEntropyWithLogitsLoss does not support string dtype");
+      throw new chunkZ6BGACIH_cjs.DTypeError("binaryCrossEntropyWithLogitsLoss does not support string dtype");
     }
     if (data instanceof BigInt64Array) {
-      const raw = chunkJSCDE774_cjs.getBigIntElement(data, loss.tensor.offset);
+      const raw = chunkZ6BGACIH_cjs.getBigIntElement(data, loss.tensor.offset);
       return Number(raw);
     }
-    return chunkJSCDE774_cjs.getNumericElement(data, loss.tensor.offset);
+    return chunkZ6BGACIH_cjs.getNumericElement(data, loss.tensor.offset);
   }
   return loss;
 }
@@ -2798,17 +2831,25 @@ function shapesEqual2(a, b) {
 }
 function ensureSameShape(a, b, context) {
   if (!shapesEqual2(a.shape, b.shape)) {
-    throw new chunkJSCDE774_cjs.ShapeError(`Shape mismatch in ${context}: [${a.shape}] vs [${b.shape}]`);
+    throw new chunkZ6BGACIH_cjs.ShapeError(`Shape mismatch in ${context}: [${a.shape}] vs [${b.shape}]`);
+  }
+}
+function alignShapes(a, b) {
+  if (shapesEqual2(a.shape, b.shape)) return [a, b];
+  if (a.size === b.size) {
+    if (a.ndim > b.ndim) return [chunk724CXHFH_cjs.reshape(a, b.shape), b];
+    if (b.ndim > a.ndim) return [a, chunk724CXHFH_cjs.reshape(b, a.shape)];
   }
+  return [a, b];
 }
 function ensureNumeric(t, context) {
   if (t.dtype === "string") {
-    throw new chunkJSCDE774_cjs.DTypeError(`${context} does not support string dtype`);
+    throw new chunkZ6BGACIH_cjs.DTypeError(`${context} does not support string dtype`);
   }
 }
 function validateReduction(reduction, context) {
   if (reduction !== "mean" && reduction !== "sum" && reduction !== "none") {
-    throw new chunkJSCDE774_cjs.InvalidParameterError(
+    throw new chunkZ6BGACIH_cjs.InvalidParameterError(
       `${context} reduction must be 'mean', 'sum', or 'none'`,
       "reduction",
       reduction
@@ -2816,38 +2857,51 @@ function validateReduction(reduction, context) {
   }
 }
 function readNumericFlat(data, flat, logicalStrides, strides, offset) {
-  const dataOffset = chunk6AE5FKKQ_cjs.offsetFromFlatIndex(flat, logicalStrides, strides, offset);
-  return chunkJSCDE774_cjs.getElementAsNumber(data, dataOffset);
+  const dataOffset = chunk724CXHFH_cjs.offsetFromFlatIndex(flat, logicalStrides, strides, offset);
+  return chunkZ6BGACIH_cjs.getElementAsNumber(data, dataOffset);
 }
 function mseLoss(predictions, targets, reduction = "mean") {
   validateReduction(reduction, "mseLoss");
-  ensureNumeric(predictions, "mseLoss");
-  ensureNumeric(targets, "mseLoss");
-  ensureSameShape(predictions, targets, "mseLoss");
-  const diff = chunk6AE5FKKQ_cjs.sub(predictions, targets);
-  const squaredDiff = chunk6AE5FKKQ_cjs.pow(diff, chunk6AE5FKKQ_cjs.tensor(2, { dtype: diff.dtype, device: diff.device }));
+  if (chunk724CXHFH_cjs.GradTensor.isGradTensor(predictions)) {
+    const pred = predictions;
+    const tgt = chunk724CXHFH_cjs.GradTensor.isGradTensor(targets) ? targets : chunk724CXHFH_cjs.GradTensor.fromTensor(targets, { requiresGrad: false });
+    const diff2 = pred.sub(tgt);
+    const squared = diff2.mul(diff2);
+    if (reduction === "none") return squared;
+    if (reduction === "sum") return squared.sum();
+    return squared.mean();
+  }
+  let preds = predictions;
+  let tgts = chunk724CXHFH_cjs.GradTensor.isGradTensor(targets) ? targets.tensor : targets;
+  ensureNumeric(preds, "mseLoss");
+  ensureNumeric(tgts, "mseLoss");
+  [preds, tgts] = alignShapes(preds, tgts);
+  ensureSameShape(preds, tgts, "mseLoss");
+  const diff = chunk724CXHFH_cjs.sub(preds, tgts);
+  const squaredDiff = chunk724CXHFH_cjs.pow(diff, chunk724CXHFH_cjs.tensor(2, { dtype: diff.dtype, device: diff.device }));
   if (reduction === "none") {
     return squaredDiff;
   }
   if (reduction === "sum") {
-    return chunk6AE5FKKQ_cjs.sum(squaredDiff);
+    return chunk724CXHFH_cjs.sum(squaredDiff);
   }
-  return chunk6AE5FKKQ_cjs.mean(squaredDiff);
+  return chunk724CXHFH_cjs.mean(squaredDiff);
 }
 function maeLoss(predictions, targets, reduction = "mean") {
   validateReduction(reduction, "maeLoss");
   ensureNumeric(predictions, "maeLoss");
   ensureNumeric(targets, "maeLoss");
+  [predictions, targets] = alignShapes(predictions, targets);
   ensureSameShape(predictions, targets, "maeLoss");
-  const diff = chunk6AE5FKKQ_cjs.sub(predictions, targets);
-  const absDiff = chunk6AE5FKKQ_cjs.abs(diff);
+  const diff = chunk724CXHFH_cjs.sub(predictions, targets);
+  const absDiff = chunk724CXHFH_cjs.abs(diff);
   if (reduction === "none") {
     return absDiff;
   }
   if (reduction === "sum") {
-    return chunk6AE5FKKQ_cjs.sum(absDiff);
+    return chunk724CXHFH_cjs.sum(absDiff);
   }
-  return chunk6AE5FKKQ_cjs.mean(absDiff);
+  return chunk724CXHFH_cjs.mean(absDiff);
 }
 function binaryCrossEntropyLoss(predictions, targets, reduction = "mean") {
   validateReduction(reduction, "binaryCrossEntropyLoss");
@@ -2855,50 +2909,51 @@ function binaryCrossEntropyLoss(predictions, targets, reduction = "mean") {
   ensureNumeric(targets, "binaryCrossEntropyLoss");
   ensureSameShape(predictions, targets, "binaryCrossEntropyLoss");
   const epsilon = 1e-7;
-  const predClamped = chunk6AE5FKKQ_cjs.clip(predictions, epsilon, 1 - epsilon);
-  const logPred = chunk6AE5FKKQ_cjs.log(predClamped);
-  const term1 = chunk6AE5FKKQ_cjs.mul(targets, logPred);
-  const one = chunk6AE5FKKQ_cjs.tensor(1, {
+  const predClamped = chunk724CXHFH_cjs.clip(predictions, epsilon, 1 - epsilon);
+  const logPred = chunk724CXHFH_cjs.log(predClamped);
+  const term1 = chunk724CXHFH_cjs.mul(targets, logPred);
+  const one = chunk724CXHFH_cjs.tensor(1, {
     dtype: predictions.dtype === "float64" ? "float64" : "float32",
     device: predictions.device
   });
-  const oneMinusTargets = chunk6AE5FKKQ_cjs.sub(one, targets);
-  const oneMinusPred = chunk6AE5FKKQ_cjs.sub(one, predClamped);
-  const logOneMinusPred = chunk6AE5FKKQ_cjs.log(oneMinusPred);
-  const term2 = chunk6AE5FKKQ_cjs.mul(oneMinusTargets, logOneMinusPred);
-  const loss = chunk6AE5FKKQ_cjs.neg(chunk6AE5FKKQ_cjs.add(term1, term2));
+  const oneMinusTargets = chunk724CXHFH_cjs.sub(one, targets);
+  const oneMinusPred = chunk724CXHFH_cjs.sub(one, predClamped);
+  const logOneMinusPred = chunk724CXHFH_cjs.log(oneMinusPred);
+  const term2 = chunk724CXHFH_cjs.mul(oneMinusTargets, logOneMinusPred);
+  const loss = chunk724CXHFH_cjs.neg(chunk724CXHFH_cjs.add(term1, term2));
   if (reduction === "none") {
     return loss;
   }
   if (reduction === "sum") {
-    return chunk6AE5FKKQ_cjs.sum(loss);
+    return chunk724CXHFH_cjs.sum(loss);
   }
-  return chunk6AE5FKKQ_cjs.mean(loss);
+  return chunk724CXHFH_cjs.mean(loss);
 }
 function rmseLoss(predictions, targets) {
   ensureNumeric(predictions, "rmseLoss");
   ensureNumeric(targets, "rmseLoss");
   ensureSameShape(predictions, targets, "rmseLoss");
   const mse = mseLoss(predictions, targets, "mean");
-  return chunk6AE5FKKQ_cjs.sqrt(mse);
+  return chunk724CXHFH_cjs.sqrt(mse);
 }
 function huberLoss(predictions, targets, delta = 1, reduction = "mean") {
   validateReduction(reduction, "huberLoss");
   ensureNumeric(predictions, "huberLoss");
   ensureNumeric(targets, "huberLoss");
+  [predictions, targets] = alignShapes(predictions, targets);
   ensureSameShape(predictions, targets, "huberLoss");
   if (!Number.isFinite(delta) || delta <= 0) {
-    throw new chunkJSCDE774_cjs.InvalidParameterError(`delta must be positive; got ${delta}`, "delta", delta);
+    throw new chunkZ6BGACIH_cjs.InvalidParameterError(`delta must be positive; got ${delta}`, "delta", delta);
   }
-  const diff = chunk6AE5FKKQ_cjs.sub(predictions, targets);
-  const absDiff = chunk6AE5FKKQ_cjs.abs(diff);
+  const diff = chunk724CXHFH_cjs.sub(predictions, targets);
+  const absDiff = chunk724CXHFH_cjs.abs(diff);
   const absData = absDiff.data;
   if (Array.isArray(absData)) {
-    throw new chunkJSCDE774_cjs.DTypeError("huberLoss does not support string dtype");
+    throw new chunkZ6BGACIH_cjs.DTypeError("huberLoss does not support string dtype");
   }
   const dtype = predictions.dtype === "float64" ? "float64" : "float32";
   const lossData = dtype === "float64" ? new Float64Array(diff.size) : new Float32Array(diff.size);
-  const logicalStrides = chunk6AE5FKKQ_cjs.computeStrides(absDiff.shape);
+  const logicalStrides = chunk724CXHFH_cjs.computeStrides(absDiff.shape);
   for (let i = 0; i < diff.size; i++) {
     const absVal = readNumericFlat(absData, i, logicalStrides, absDiff.strides, absDiff.offset);
     if (absVal <= delta) {
@@ -2907,7 +2962,7 @@ function huberLoss(predictions, targets, delta = 1, reduction = "mean") {
       lossData[i] = delta * (absVal - 0.5 * delta);
     }
   }
-  const loss = chunk6AE5FKKQ_cjs.Tensor.fromTypedArray({
+  const loss = chunk724CXHFH_cjs.Tensor.fromTypedArray({
     data: lossData,
     shape: predictions.shape,
     dtype,
@@ -2917,9 +2972,9 @@ function huberLoss(predictions, targets, delta = 1, reduction = "mean") {
     return loss;
   }
   if (reduction === "sum") {
-    return chunk6AE5FKKQ_cjs.sum(loss);
+    return chunk724CXHFH_cjs.sum(loss);
   }
-  return chunk6AE5FKKQ_cjs.mean(loss);
+  return chunk724CXHFH_cjs.mean(loss);
 }
 
 exports.AvgPool2d = AvgPool2d;
@@ -2956,5 +3011,5 @@ exports.maeLoss = maeLoss;
 exports.mseLoss = mseLoss;
 exports.nn_exports = nn_exports;
 exports.rmseLoss = rmseLoss;
-//# sourceMappingURL=chunk-QERHVCHC.cjs.map
-//# sourceMappingURL=chunk-QERHVCHC.cjs.map
+//# sourceMappingURL=chunk-EUZHJDZ6.cjs.map
+//# sourceMappingURL=chunk-EUZHJDZ6.cjs.map