Support for attention bias in gemma + refactor things a bit. (#1744)

* Support for attention bias in gemma + refactor things a bit.

* Fix the cuda tests.

1 files changed, 1 insertions, 9 deletions

diff --git a/candle-transformers/src/models/chatglm.rs b/candle-transformers/src/models/chatglm.rs
index 95466b34..0686b34e 100644
--- a/candle-transformers/src/models/chatglm.rs
+++ b/candle-transformers/src/models/chatglm.rs
@@ -1,4 +1,4 @@
-use crate::models::with_tracing::Linear;
+use crate::models::with_tracing::{linear_b as linear, Linear};
 use candle::{DType, Device, IndexOp, Module, Result, Tensor, D};
 use candle_nn::VarBuilder;
 
@@ -51,14 +51,6 @@ impl Config {
     }
 }
 
-fn linear(in_dim: usize, out_dim: usize, bias: bool, vb: VarBuilder) -> Result<Linear> {
-    if bias {
-        crate::models::with_tracing::linear(in_dim, out_dim, vb)
-    } else {
-        crate::models::with_tracing::linear_no_bias(in_dim, out_dim, vb)
-    }
-}
-
 #[derive(Debug, Clone)]
 struct RotaryEmbedding {
     cache: Tensor,