fix fused_attention with mp unit test case fail on A100 with CUDA >= 11.6 (#45883)

a5836222 · LiYuRio · GitHub · 7c1dc754 · a5836222
隐藏空白更改
内联并排

Showing with 1 addition and 1 deletion

python/paddle/fluid/tests/unittests/static_model_parallel_fused_attention.py .../tests/unittests/static_model_parallel_fused_attention.py +1 -1

未找到文件。
--- a/python/paddle/fluid/tests/unittests/static_model_parallel_fused_attention.py
+++ b/python/paddle/fluid/tests/unittests/static_model_parallel_fused_attention.py
@@ -36,7 +36,7 @@ def get_param_attr(weight, bias):
 DTYPE = "float32"
 MODEL_PARALLEL_SIZE = 2
 n_head = 2 * MODEL_PARALLEL_SIZE
-d_key = 4
+d_key = 2
 hidden = n_head * d_key