100+ Free TensorFlow Developer Practice Questions

Pass your Google TensorFlow Developer Certificate exam on the first try — instant access, no signup required.

✓ No registration✓ No credit card✓ No hidden fees✓ Start practicing immediately

~60-70% Pass Rate

100+ Questions

100% Free

1 / 10

Question 1

Score: 0/0

Which Keras API is the simplest way to build a plain feedforward neural network with a linear stack of layers?

tf.keras.Sequential

tf.keras.Model (Functional API)

tf.keras.layers.Subclassing

tf.estimator

to track

2026 Statistics

Key Facts: TensorFlow Developer Exam

Coding Tasks

Google

90%

Passing Score

Google (min 3/5 per task)

5 hours

Exam Duration

Google

$100

Exam Fee

Google

PyCharm

IDE

TensorFlow Exam plugin

3 years

Validity

Must retake

The TensorFlow Developer Certificate costs $100 and is a 5-hour coding exam in PyCharm. Candidates build five models (regression, computer vision, NLP, time-series, mixed) using TensorFlow 2.x and Keras, save them, and submit for auto-grading. Pass threshold is 90% overall, minimum 3/5 per task. Certificate valid 3 years.

Sample TensorFlow Developer Practice Questions

Try these sample questions to test your TensorFlow Developer exam readiness. Each question includes a detailed explanation. Start the interactive quiz above for the full 100+ question experience with AI tutoring.

1Which Keras API is the simplest way to build a plain feedforward neural network with a linear stack of layers?

A.tf.keras.Sequential

B.tf.keras.Model (Functional API)

C.tf.keras.layers.Subclassing

D.tf.estimator

Explanation: tf.keras.Sequential stacks layers linearly and is the simplest API — ideal for straightforward architectures. The Functional API is used for multi-input/output or branching models. Subclassing is for full flexibility. tf.estimator is a legacy high-level API largely superseded by Keras.

2Which layer flattens a multi-dimensional input (e.g., 28x28 image) into a 1D vector before a Dense layer?

A.tf.keras.layers.Reshape

B.tf.keras.layers.Flatten

C.tf.keras.layers.Dense(flatten=True)

D.tf.keras.layers.GlobalAveragePooling2D

Explanation: Flatten reshapes (batch, h, w, c) into (batch, h*w*c) — the canonical pre-Dense layer for simple image models. Reshape can do this but requires explicit shape. Dense has no flatten arg. GlobalAveragePooling2D pools spatially to (batch, c), not a flatten.

3Which Keras method trains a model on training data?

A.model.train()

B.model.fit()

C.model.run()

D.model.learn()

Explanation: model.fit(x, y, epochs=..., batch_size=...) trains a compiled Keras model. model.evaluate computes test metrics; model.predict runs inference. train(), run(), and learn() are not Keras methods.

4Which loss function is appropriate for multi-class classification with integer labels (e.g., y=3)?

A.binary_crossentropy

B.categorical_crossentropy

C.sparse_categorical_crossentropy

D.mean_squared_error

Explanation: sparse_categorical_crossentropy expects integer labels and one softmax probability vector per sample. Use categorical_crossentropy with one-hot encoded labels. binary_crossentropy is for two-class/multi-label. MSE is for regression, not classification.

5What does this code configure? model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

A.A regression model with MSE loss

B.A binary classification model using Adam optimizer tracking accuracy

C.A multi-class classification model with 10 classes

D.An autoencoder

Explanation: binary_crossentropy is the standard loss for binary classification (sigmoid output) or multi-label classification. Adam is an adaptive optimizer. metrics=['accuracy'] computes classification accuracy during training and evaluation.

6Which layer is the core convolutional layer for 2D image processing?

A.tf.keras.layers.Conv1D

B.tf.keras.layers.Conv2D

C.tf.keras.layers.Conv3D

D.tf.keras.layers.Dense

Explanation: Conv2D applies 2D convolution (a 2D filter sliding over height and width), the standard building block for image CNNs. Conv1D is for sequences. Conv3D is for volumetric data (video, medical). Dense is fully connected.

7Which layer reduces spatial dimensions by taking the maximum value in each pooling window?

A.AveragePooling2D

B.MaxPooling2D

C.BatchNormalization

D.Dropout

Explanation: MaxPooling2D takes the max value in each window, downsampling spatial dimensions and providing translation invariance. AveragePooling2D takes the mean. BatchNormalization normalizes activations. Dropout randomly zeros units for regularization.

8Which data augmentation layer randomly flips images horizontally during training?

A.tf.keras.layers.RandomFlip('horizontal')

B.tf.keras.layers.Flip('horizontal')

C.tf.keras.layers.ImageFlip('h')

D.tf.keras.layers.HFlip()

Explanation: tf.keras.layers.RandomFlip('horizontal') is the modern Keras preprocessing layer that flips images left-right with 50% probability during training only (pass-through at inference). Other options are not valid Keras layers.

9Which tf.keras function loads images from a directory tree organized by class subfolder?

A.tf.keras.utils.image_dataset_from_directory

B.tf.data.Dataset.from_directory

C.tf.io.read_image_directory

D.ImageDataGenerator.fit()

Explanation: image_dataset_from_directory reads a folder tree like class_a/xxx.jpg, class_b/yyy.jpg and returns a tf.data.Dataset of (image, label) batches. It replaces the older ImageDataGenerator.flow_from_directory. The other options are either non-existent or incorrect usage.

10Which tf.keras.applications model is a lightweight architecture designed for mobile/edge devices?

A.VGG16

B.MobileNetV2

C.InceptionV3

D.ResNet152

Explanation: MobileNetV2 uses depthwise-separable convolutions and inverted residual blocks to be lightweight and fast on mobile/edge. VGG16, InceptionV3, and ResNet152 are larger and targeted at server-class inference. EfficientNet-Lite is another mobile option.

About the TensorFlow Developer Exam

The TensorFlow Developer Certificate is a performance-based exam from Google that validates the ability to build deep learning models using TensorFlow 2.x and the Keras API. The exam is taken in a dedicated PyCharm environment and consists of five coding tasks: basic regression, image classification with CNNs, natural language processing with embeddings and RNNs, time-series forecasting, and a mixed real-world problem.

Questions

5 scored questions

Time Limit

5 hours

Passing Score

90% overall (minimum 3/5 per task)

Exam Fee

$100 (Google / Trueability (PSI))

TensorFlow Developer Exam Content Outline

15-20%

TensorFlow & Keras Fundamentals

TensorFlow 2.x basics, eager execution, tf.function, Sequential/Functional/Subclassing API, model.compile, model.fit, optimizers (Adam, SGD, RMSprop), losses, metrics

20-25%

Image Classification with CNNs

Conv2D, MaxPooling2D, Flatten/GlobalAveragePooling, data augmentation (RandomFlip, RandomRotation), image_dataset_from_directory, transfer learning with tf.keras.applications, fine-tuning

20-25%

Natural Language Processing

TextVectorization, Embedding layer, LSTM/GRU/Bidirectional RNNs, sentiment analysis, mask_zero, padding sequences, subword tokenization

15-20%

Time-Series Forecasting

Windowed datasets, batching sequences, Conv1D, LSTM, RNN for forecasting, multivariate series, Huber loss, learning rate scheduling

15-20%

Data Pipelines, Saving & Deployment

tf.data Dataset API (map, batch, shuffle, prefetch with AUTOTUNE), callbacks (EarlyStopping, ReduceLROnPlateau, ModelCheckpoint), saving/loading models (SavedModel, HDF5), TFLite conversion

How to Pass the TensorFlow Developer Exam

What You Need to Know

Passing score: 90% overall (minimum 3/5 per task)
Exam length: 5 questions
Time limit: 5 hours
Exam fee: $100

Keys to Passing

Complete 500+ practice questions
Score 80%+ consistently before scheduling
Focus on highest-weighted sections
Use our AI tutor for tough concepts

TensorFlow Developer Study Tips from Top Performers

1Install PyCharm and the TensorFlow Exam plugin early — the exam only works in this environment

2For image tasks, default to a small Conv2D stack with BatchNormalization and Dropout; know transfer learning with MobileNetV2/EfficientNet

3For NLP, use TextVectorization with an Embedding(mask_zero=True) and a Bidirectional(LSTM) or Conv1D

4For time-series, build windowed tf.data datasets and use Huber loss with Conv1D + LSTM

5Use callbacks — EarlyStopping(patience), ReduceLROnPlateau, ModelCheckpoint(save_best_only=True)

6Always save the final model with model.save('mymodel.h5') — that is what the grader checks

7Test CPU inference time — exam submissions are graded on a CPU; avoid huge models that time out

Frequently Asked Questions

What is the TensorFlow Developer Certificate?

The TensorFlow Developer Certificate is a performance-based credential from Google that validates a developer's ability to build, train, and deploy deep learning models using TensorFlow 2.x and the Keras API. Unlike multiple-choice exams, it requires you to write code to solve five ML problems.

How is the TensorFlow Developer exam structured?

The exam is a 5-hour online coding test administered through a dedicated PyCharm IDE plugin. You solve five tasks: a basic regression, image classification with a CNN, NLP/sentiment analysis with an RNN, time-series forecasting, and a general real-world dataset problem. Each task requires you to build a Keras model, train it, and submit it.

What is the passing score for the TensorFlow Developer exam?

You must achieve 90% overall AND score at least 3 out of 5 on every individual task. Even if your overall score is high, a single task scoring below 3/5 will fail you. This design forces balanced skill across regression, vision, NLP, and time-series.

How much does the TensorFlow Developer Certificate cost?

The exam fee is $100 USD. Google provides a TensorFlow education stipend program for students and candidates who cannot afford the fee. You have 6 months from registration to take the exam. If you fail, retake policies vary (typically 14-day then 2-month waits).

How should I prepare for the TensorFlow Developer exam?

Plan for 50-80 hours. Complete the DeepLearning.AI 'TensorFlow in Practice' specialization on Coursera — it was designed as the official prep. Practice Conv2D stacks for images, Embedding + LSTM for text, and windowed datasets for time-series. Memorize common Keras boilerplate since you'll re-type it on exam day.

How long is the TensorFlow Developer Certificate valid?

The certificate is valid for 3 years. After that, you must retake the exam to maintain active status. Note: As of 2024 Google paused new registrations while evaluating the next iteration — verify current availability on the official TensorFlow certificate page.