Deep Learning and PyTorch Bootcamp for Beginners (Easy! From Basics to ChatGPT's Core Transformer) [Data Analysis/Science Part 3]
This is a newly designed course that allows you to gradually learn the mathematics, theory, PyTorch-based implementation, transfer learning, and GPT's core transformer needed to understand deep learning, based on the instructor's own failed experiences when first learning deep learning.
1,676 learners
Level Basic
Course period Unlimited
Sharing changes to the version of PyTorch in the first deep learning and PyTorch bootcamp
Hello. This is Dave Lee from Janjaemi Coding.
The reason is that I am sharing this because there are some parts of the code that may behave slightly differently in recent PyTorch versions.
In recent PyTorch versions, when creating a tensor with FloatTensor() , each tensor is initialized to 0, not a random value.
(Because the internal memory is allocated but the values are not initialized to “0” or random values, random-looking values (actually garbage values that were left in the memory) were entered, but the latest version forces them to be initialized to 0.)
Therefore, when creating a simple tensor without arbitrarily specifying a specific value, 0 is specified as the initial value when outputting each tensor, not a random value, so it comes out as 0 when output. This is normal operation, so it would be good to refer to only this part.
We have updated the entire data by including an explanation of this part in the code data today.
thank you
Dave Lee Dream of residual fun coding