Fix typos in the documentation strings of the tfjs-layers directory (#…

…8411)

tfjs-layers/README.md

@@ -52,7 +52,7 @@ const ys = tf.tensor2d([[1], [3], [5], [7]], [4, 1]);
 // Train the model.
 await model.fit(xs, ys, {epochs: 500});
 
-// Ater the training, perform inference.
+// After the training, perform inference.
 const output = model.predict(tf.tensor2d([[5]], [1, 1]));
 output.print();
 ```

tfjs-layers/demos/README.md

@@ -16,7 +16,7 @@ Once the development environment is prepared, execute the build script from the
 ```
 
 The script will construct a number of Keras models in Python and benchmark their training using the TensorFlow backend.  When it is complete, it will bring up a
-local HTTP server.  Navigate to the local URL spcecified in stdout to bring up
+local HTTP server.  Navigate to the local URL specified in stdout to bring up
 the benchmarks page UI.  There will be a button to begin the JS side of the
 benchmarks.  Clicking the button will run through and time the same models, now
 running in the browser.

tfjs-layers/src/layers/nlp/modeling/transformer_decoder_test.ts

@@ -106,7 +106,7 @@ describe('TransformerDecoder', () => {
     const config = testLayer.getConfig();
     const restored = TransformerDecoder.fromConfig(TransformerDecoder, config);
 
-    // Initializers don't get serailized with customObjects.
+    // Initializers don't get serialized with customObjects.
     delete ((config['kernelInitializer'] as serialization.ConfigDict
       )['config'] as serialization.ConfigDict)['customObjects'];
     delete ((config['biasInitializer'] as serialization.ConfigDict
@@ -167,5 +167,5 @@ describe('TransformerDecoder', () => {
     expectTensorsClose(outputCache, noLoopCache);
   });
 
-  // TODO(pforderique): Test mask propogation once supported.
+  // TODO(pforderique): Test mask propagation once supported.
 });

tfjs-layers/src/layers/nlp/models/gpt2/gpt2_causal_lm.ts

@@ -70,7 +70,7 @@ export declare interface GPT2CausalLMArgs extends PipelineModelArgs {
 }
 
 /**
- * An end-to-end GPT2 model for causal langauge modeling.
+ * An end-to-end GPT2 model for causal language modeling.
  *
  * A causal language model (LM) predicts the next token based on previous
  * tokens. This task setup can be used to train the model unsupervised on

tfjs-layers/src/layers/nlp/multihead_attention.ts

@@ -703,7 +703,7 @@ export class MultiHeadAttention extends Layer {
 
     newInputs = [inputs, kwargs['value']].concat(kwargs['key'] ?? []);
 
-    // TODO(pforderique): Support mask propogation.
+    // TODO(pforderique): Support mask propagation.
     return super.apply(newInputs, kwargs);
   }
 

tfjs-layers/src/layers/normalization.ts

@@ -430,7 +430,7 @@ export interface LayerNormalizationLayerArgs extends LayerArgs {
   axis?: number|number[];
 
   /**
-   * A small positive float added to variance to avoid divison by zero.
+   * A small positive float added to variance to avoid division by zero.
    * Defaults to 1e-3.
    */
   epsilon?: number;

tfjs-layers/src/layers/normalization_test.ts

@@ -353,7 +353,7 @@ describeMathCPUAndWebGL2('BatchNormalization Layers: Tensor', () => {
     const x = tensor2d([[1, 2], [3, 4]], [2, 2]);
     expectTensorsClose(layer.apply(x) as Tensor, x, 0.01);
     expect(layer.getWeights().length).toEqual(3);
-    // Firt weight is gamma.
+    // First weight is gamma.
     expectTensorsClose(layer.getWeights()[0], onesLike(layer.getWeights()[0]));
     // Second weight is moving mean.
     expectTensorsClose(layer.getWeights()[1], zerosLike(layer.getWeights()[1]));
@@ -366,7 +366,7 @@ describeMathCPUAndWebGL2('BatchNormalization Layers: Tensor', () => {
     const x = tensor2d([[1, 2], [3, 4]], [2, 2]);
     expectTensorsClose(layer.apply(x) as Tensor, x, 0.01);
     expect(layer.getWeights().length).toEqual(3);
-    // Firt weight is beta.
+    // First weight is beta.
     expectTensorsClose(layer.getWeights()[0], zerosLike(layer.getWeights()[0]));
     // Second weight is moving mean.
     expectTensorsClose(layer.getWeights()[1], zerosLike(layer.getWeights()[1]));

tfjs-layers/src/layers/preprocessing/image_resizing_test.ts

@@ -88,7 +88,7 @@ describeMathCPUAndGPU('Resizing Layer', () => {
   });
 
   it('Returns a tensor of the correct dtype', () => {
-    // do a same resizing operation, cheeck tensors dtypes and content
+    // do a same resizing operation, check tensors dtypes and content
     const height = 40;
     const width = 60;
     const numChannels = 3;

tfjs-layers/src/layers/preprocessing/random_height.ts

@@ -35,7 +35,7 @@ type InterpolationType = typeof INTERPOLATION_KEYS[number];
  *
  * The input should be a 3D (unbatched) or
  * 4D (batched) tensor in the `"channels_last"` image data format. Input pixel
- * values can be of any range (e.g. `[0., 1.)` or `[0, 255]`) and of interger
+ * values can be of any range (e.g. `[0., 1.)` or `[0, 255]`) and of integer
  * or floating point dtype. By default, the layer will output floats.
  *
  * tf methods implemented in tfjs: 'bilinear', 'nearest',
@@ -48,7 +48,7 @@ export class RandomHeight extends BaseRandomLayer {
   /** @nocollapse */
   static override className = 'RandomHeight';
   private readonly factor: number | [number, number];
-  private readonly interpolation?: InterpolationType;  // defualt = 'bilinear
+  private readonly interpolation?: InterpolationType;  // default = 'bilinear
   private heightLower: number;
   private heightUpper: number;
   private imgWidth: number;

tfjs-layers/src/layers/preprocessing/random_width.ts

@@ -35,7 +35,7 @@ type InterpolationType = typeof INTERPOLATION_KEYS[number];
  *
  * The input should be a 3D (unbatched) or
  * 4D (batched) tensor in the `"channels_last"` image data format. Input pixel
- * values can be of any range (e.g. `[0., 1.)` or `[0, 255]`) and of interger
+ * values can be of any range (e.g. `[0., 1.)` or `[0, 255]`) and of integer
  * or floating point dtype. By default, the layer will output floats.
  *
  * tf methods implemented in tfjs: 'bilinear', 'nearest',
@@ -48,7 +48,7 @@ export class RandomWidth extends BaseRandomLayer {
   /** @nocollapse */
   static override className = 'RandomWidth';
   private readonly factor: number | [number, number];
-  private readonly interpolation?: InterpolationType;  // defualt = 'bilinear
+  private readonly interpolation?: InterpolationType;  // default = 'bilinear
   private widthLower: number;
   private widthUpper: number;
   private imgHeight: number;

tfjs-layers/src/layers/recurrent.ts

@@ -252,7 +252,7 @@ export declare interface BaseRNNLayerArgs extends LayerArgs {
    *     see section "Note on passing external constants" below.
    *     Porting Node: PyKeras overrides the `call()` signature of RNN cells,
    *       which are Layer subtypes, to accept two arguments. tfjs-layers does
-   *       not do such overriding. Instead we preseve the `call()` signature,
+   *       not do such overriding. Instead we preserve the `call()` signature,
    *       which due to its `Tensor|Tensor[]` argument and return value is
    *       flexible enough to handle the inputs and states.
    *   - a `stateSize` attribute. This can be a single integer (single state)
@@ -757,7 +757,7 @@ export class RNN extends Layer {
 
       const output = this.returnSequences ? outputs : lastOutput;
 
-      // TODO(cais): Porperty set learning phase flag.
+      // TODO(cais): Property set learning phase flag.
 
       if (this.returnState) {
         return [output].concat(states);
@@ -1933,7 +1933,7 @@ export class StackedRNNCells extends RNNCell {
 
   get stateSize(): number[] {
     // States are a flat list in reverse order of the cell stack.
-    // This allows perserving the requirement `stack.statesize[0] ===
+    // This allows preserving the requirement `stack.statesize[0] ===
     // outputDim`. E.g., states of a 2-layer LSTM would be `[h2, c2, h1, c1]`,
     // assuming one LSTM has states `[h, c]`.
     const stateSize: number[] = [];
@@ -2098,7 +2098,7 @@ export class StackedRNNCells extends RNNCell {
     batchSetValue(tuples);
   }
 
-  // TODO(cais): Maybe implemnt `losses` and `getLossesFor`.
+  // TODO(cais): Maybe implement `losses` and `getLossesFor`.
 }
 serialization.registerClass(StackedRNNCells);