Clarify shape descriptions inside forward method #9
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
you're right, i like it, ty |
|
This is my very first merged PR to an open source project. Feels nice 🙂 And I want to use my 3 seconds of fame to say big thank you for the videos that you've made so far and I hope you are not planning to stop. |
huiqiao
pushed a commit
to huiqiao/char-level-gpt
that referenced
this pull request
Mar 15, 2024
…ription_fix Clarify shape descriptions inside forward method
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I think that shape descriptions in forward method are somewhat misleading.
The input has size (B, T, C) and then this
Cis used to describe shape on every step during forward pass, where in fact the last dimension is actuallyhead_size.Ccan be equal tohead_size(if for example the number of heads is 1), but it's not guaranteed.In addition the shape of projection layer should have
(head_size * num_heads, n_embd)as input and output dimensions respectively. This will make it more robust. In this code the size of head is equal ton_embd // num_heads, but it's not a strict rule and we might want to not reduce the size of a single head (by a factor of num_heads) and as a result the last dimension of concatenated heads will be larger than embedding size. Like it's shown in this article.