Clippy

Author: EricLBuehler

mistralrs-core/src/attention.rs

@@ -386,34 +386,30 @@ impl Sdpa {
 
                 // Batch matrix multiplication
                 // Fuse softmax scale and attention_bias add
-                let mut attention_scores = cublaslt
-                    .batch_matmul(
-                        &k,
-                        &q,
-                        attention_bias.as_ref(),
-                        Some(sdpa_params.softmax_scale / sdpa_params.softcap.unwrap_or(1.0)),
-                        beta,
-                        None,
-                        None,
-                    )
-                    .unwrap();
+                let mut attention_scores = cublaslt.batch_matmul(
+                    &k,
+                    &q,
+                    attention_bias.as_ref(),
+                    Some(sdpa_params.softmax_scale / sdpa_params.softcap.unwrap_or(1.0)),
+                    beta,
+                    None,
+                    None,
+                )?;
                 if let Some(softcap) = sdpa_params.softcap {
                     attention_scores = (attention_scores.tanh()? * softcap as f64)?;
                 }
                 candle_nn::ops::inplace_softmax_last_dim(&mut attention_scores)?;
 
-                let context_layer = cublaslt
-                    .batch_matmul(
-                        &v.t()?.contiguous().unwrap(),
-                        &attention_scores,
-                        // We save one allocation
-                        Some(&q),
-                        None,
-                        None,
-                        None,
-                        None,
-                    )
-                    .unwrap();
+                let context_layer = cublaslt.batch_matmul(
+                    &v.t()?.contiguous()?,
+                    &attention_scores,
+                    // We save one allocation
+                    Some(&q),
+                    None,
+                    None,
+                    None,
+                    None,
+                )?;
 
                 // Reshape to dims4
                 context_layer.reshape((b_sz, n_attn_heads, seq_len, v_head_dim))