Skip to content

fix: split input sequence correctly and enable using kv_dim#116

Open
richardk53 wants to merge 1 commit intomainfrom
fix-perceiver-kv-dim
Open

fix: split input sequence correctly and enable using kv_dim#116
richardk53 wants to merge 1 commit intomainfrom
fix-perceiver-kv-dim

Conversation

@richardk53
Copy link

Previously, kv_dim has not been enabled together with modulation and instead thrown an assert.
This is because the input sequence is simply split into 8 equally sized parts, which is incorrect if kv_dim differs.
This PR fixes this.

I have read the CLA Document and I hereby sign the CLA

… parts for kv and 6 parts for q instead of 8 same-sized parts (which is wrong if kv_dim is different)
@github-actions
Copy link

Coverage

Tests Skipped Failures Errors Time
1099 21 💤 0 ❌ 0 🔥 30.460s ⏱️

Copy link
Collaborator

@MauritsBleeker MauritsBleeker left a comment

Choose a reason for hiding this comment

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

Can/should we add a test for this?

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