Skip to content

Conversation

davidbrandfonbrener
Copy link
Contributor

The k_norm should be the dimension of the key vector, which is head_dim * effective_n_kv_heads, and not d_model / effective_n_kv_heads.

Also fixing unnecessary assert.

The k_norm should be the dimension of the key vector, which is head_dim * effective_n_kv_heads, and not d_model / effective_n_kv_heads.

Also fixing unnecessary assert.
Copy link
Member

@dirkgr dirkgr left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks! This fell through the cracks when we implemented GQA.

@davidbrandfonbrener davidbrandfonbrener merged commit 8472d0b into main Mar 26, 2024
@davidbrandfonbrener davidbrandfonbrener deleted the davidbrandfonbrener-patch-1 branch March 26, 2024 19:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants