Skip to content

Commit

Permalink
Fix for gemma MQA.
Browse files Browse the repository at this point in the history
  • Loading branch information
LaurentMazare committed Apr 19, 2024
1 parent 9c532ae commit adb1a23
Showing 1 changed file with 3 additions and 2 deletions.
5 changes: 3 additions & 2 deletions candle-transformers/src/models/gemma.rs
Original file line number Diff line number Diff line change
Expand Up @@ -227,8 +227,9 @@ impl Attention {
};
self.kv_cache = Some((key_states.clone(), value_states.clone()));

let key_states = crate::utils::repeat_kv(key_states, self.num_kv_groups)?;
let value_states = crate::utils::repeat_kv(value_states, self.num_kv_groups)?;
let key_states = crate::utils::repeat_kv(key_states, self.num_kv_groups)?.contiguous()?;
let value_states =
crate::utils::repeat_kv(value_states, self.num_kv_groups)?.contiguous()?;

let attn_output = {
let scale = 1f64 / f64::sqrt(self.head_dim as f64);
Expand Down

0 comments on commit adb1a23

Please sign in to comment.