Deep Learning Practice Quiz 2: 1 December 2024 (September 2024 term)
The IIT Madras BS Deep Learning Practice (Deep Learning Practice) Quiz 2 paper sat on 1 Dec 2024, in the September 2024 term: 15 questions for 50 marks in 120 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.
- 15
- 50
- 120 min
- 14
- 1
Show answer
Correct answer
Question 2
(2, 3, 4, 1)
(1, 3, 5, 1)
(3, 5, 1)
(3, 1, 1, 5)
Show answer
Correct answer
(3, 1, 1, 5)
Question 3
For a 1D convolution, given the following values:
- Input size (): 50
- In channels: 16
- Kernel size: 3
- Padding: 1
- Dilation: 1
- Stride: 2
- Out channels: 32
What is the shape of the output of the convolutional layer?
32 × 25
32 × 26
16 × 32 × 50
50 × 3
Show answer
Correct answer
32 × 25
Question 4
If an audio is of duration 10 seconds and sample rate is 48KHz, what happens to the length of audio in samples if the audio is downsampled to 16KHz?
Show answer
Correct answer
Question 5
If the current learning rate is 0.002 and the maximum learning rate to be achieved in 1000 steps is 0.004, what should be the step size for a linear learning rate scheduler?
0.0002
0.00002
0.0000002
0.000002
Show answer
Correct answer
0.000002
Question 6
What is the correct order of training commands?
Show answer
Correct answers
Question 7
To quantize the Whisper model to 8-bit integer during loading in Hugging Face, which of the following arguments should be passed?
Show answer
Correct answer
Question 8
Which of the following models can be used to obtain segments for speaker diarization?
Whisper
Wav2Vec
Speecht5
GPT-3
Show answer
Correct answer
Whisper
Question 9
Which of the following arguments should be passed during dataset loading to avoid loading the full dataset at once?
Show answer
Correct answer
Question 10
Which of the following loss functions can be used for speaker classification?
Show answer
Correct answer
Question 11
If a dataset has 25600 samples, batch size is 100, each epoch processes the whole dataset only once, and gradient accumulation is set to 32, how many updates does the model make?
16
256
32
8
Show answer
Correct answer
8
Question 12
Which of the following functions can be used to replace certain values in a PyTorch tensor?
Show answer
Correct answer
Question 13
Which of the following trainers is commonly used for ASR sequence-to-sequence training?
ASRTrainer
Seq2SeqTrainer
LanguageModelTrainer
ClassificationTrainer
Show answer
Correct answer
Seq2SeqTrainer
Question 14
What type of model is HiFi-GAN?
Text-to-Text model
Vocoder
Decoder
Classifier
Show answer
Correct answer
Vocoder
Question 15
Which of the following should be done when a CUDA Out of Memory error is encountered during training?
Show answer
Correct answer