Do parameter count calculations in 64 bits to not overflow in case of… #367
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
… very large models
|
Just a reminder that |
|
@kroggen thank you, updated to unsigned long long. |
|
thank you @janimo |
This was referenced Sep 1, 2023
Closed
vinhtran2611
pushed a commit
to vinhtran2611/llama2.c
that referenced
this pull request
Jan 20, 2024
Do parameter count calculations in 64 bits to not overflow in case of…
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Outdated suggestions cannot be applied.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.
… very large models
This is based on discussions and code in #111, #154 and #164 so all credits to their authors, and is tested with the 13B and 34B code llama models. It seems to be the minimal amount of conceptual change needed.
The alternative to declaring the new n_layers variable are explicit (unsigned long)p->n_layers casts in 5 of the multiplications.
The p->n_layers*p->n_dim could also be precomputed in unsigned long and reused since it appears in all multiplications, but it would lose the pedagogical value of having the order of operations reflect the order of matrices in the cases where dim comes last.