add gpu_cpu_map_matmul_to_mul_pass to kGpuLowerPrecisionPasses (#49753)

* add gpu_cpu_map_matmul_to_mul_pass to kGpuLowerPrecisionPasses * disable fc_elementwise_layernorm_fuse_pass in mixed precision

add gpu_cpu_map_matmul_to_mul_pass to kGpuLowerPrecisionPasses (#49753)
* add gpu_cpu_map_matmul_to_mul_pass to kGpuLowerPrecisionPasses * disable fc_elementwise_layernorm_fuse_pass in mixed precision
07514139 · Yuanle Liu · GitHub · 4d5265b8 · 07514139
隐藏空白更改
内联并排

Showing with 2 addition and 1 deletion

paddle/fluid/inference/api/paddle_pass_builder.cc paddle/fluid/inference/api/paddle_pass_builder.cc +2 -1

未找到文件。
--- a/paddle/fluid/inference/api/paddle_pass_builder.cc
+++ b/paddle/fluid/inference/api/paddle_pass_builder.cc
@@ -193,8 +193,9 @@ const std::vector<std::string> kGpuLowerPrecisionPasses{
    "fuse_multi_transformer_layer_pass",
    "gpu_cpu_map_matmul_v2_to_mul_pass",
    "gpu_cpu_map_matmul_v2_to_matmul_pass",
+    "gpu_cpu_map_matmul_to_mul_pass",
    "fc_fuse_pass",
-    "fc_elementwise_layernorm_fuse_pass",
+    // "fc_elementwise_layernorm_fuse_pass",
    "embedding_eltwise_layernorm_fuse_pass",
    "inplace_op_var_pass"};