暂时注释xlmroberta中的attention加速

ztxz16 · Sep 25, 2024 · 5912fd3 · 5912fd3
1 parent c9ae3fa
commit 5912fd3
Showing 1 changed file with 3 additions and 1 deletion.
diff --git a/src/models/xlmroberta.cpp b/src/models/xlmroberta.cpp
@@ -112,7 +112,9 @@ namespace fastllm {
             PermuteSelf(k, {0, 2, 1, 3});
             PermuteSelf(v, {0, 2, 1, 3});
 
-            if (bsz == 1) {
+            if (false) {
+                // TODO: 这里使用的AttentionMask不是因果Mask，无法直接调用Attention函数
+                // 后续需要修改AttentionMask使得可以直接调用Attention函数
                 q.Reshape({-1, q.dims[2], q.dims[3]});
                 k.Reshape({-1, k.dims[2], k.dims[3]});
                 v.Reshape({-1, v.dims[2], v.dims[3]});