add fuse_multi_transformer passes to fp16. test=develop (#47733)

PaddlePaddle · Nov 8, 2022 · 34f67a8 · 34f67a8
1 parent b3e4571
commit 34f67a8
Showing 1 changed file with 6 additions and 0 deletions.
diff --git a/paddle/fluid/inference/api/paddle_pass_builder.cc b/paddle/fluid/inference/api/paddle_pass_builder.cc
@@ -163,6 +163,12 @@ const std::vector<std::string> kGpuLowerPrecisionPasses{
     "conv_elementwise_add2_act_fuse_pass",
     "conv_elementwise_add_fuse_pass",
     "multihead_matmul_fuse_pass_v2",
+    "fused_multi_transformer_encoder_pass",
+    "fused_multi_transformer_decoder_pass",
+    "fused_multi_transformer_encoder_fuse_qkv_pass",
+    "fused_multi_transformer_decoder_fuse_qkv_pass",
+    "multi_devices_fused_multi_transformer_encoder_fuse_qkv_pass",
+    "multi_devices_fused_multi_transformer_decoder_fuse_qkv_pass",
     "gpu_cpu_map_matmul_v2_to_mul_pass",
     "gpu_cpu_map_matmul_v2_to_matmul_pass",
     "fc_fuse_pass",