fix typos (PaddlePaddle#76353)

* fix typos

* fix typos

Author: co63oc

paddle/fluid/framework/ir/layer_norm_fuse_pass.cc

@@ -72,7 +72,7 @@ bool validateReduceOpAttrs(const Node* node,
     }
     for (size_t i = 1; i < dims.size(); ++i) {
       if (1 != dims[i] - dims[i - 1]) {
-        LOG(WARNING) << "The LayerNorm dim of mean must be  continuous";
+        LOG(WARNING) << "The LayerNorm dim of mean must be continuous";
         return false;
       }
     }

paddle/fluid/framework/ir/preln_residual_bias_fuse_pass.cc

@@ -217,7 +217,7 @@ int PrelnResidualBiasFusePass::ApplyPattern(ir::Graph *graph,
     // elementwise1_out. This will lead to two or more PrelnResidualBias
     // patterns is found near elementwise1_out, and these patterns will interact
     // on each other, so we make below check to ensure only one
-    // PrelnResidualBias pattern is delalted with.
+    // PrelnResidualBias pattern is dealt with.
     for (auto op : elementwise1_out->inputs) {
       if (op->Name() == "fused_bias_dropout_residual_layer_norm") return;
     }

paddle/fluid/framework/ir/preln_skip_layernorm_fuse_pass.cc

@@ -117,8 +117,7 @@ void PrelnSkipLayerNormFusePass::ApplyImpl(ir::Graph *graph) const {
                "use_varseqlen, preln_embedding_eltwise_layernorm_fuse_pass, "
                "trt_multihead_matmul_fuse_pass, "
                "set pos_id, set mask_id, with_dynamic_shape. Stop this pass, "
-               "please "
-               "reconfig.";
+               "please reconfig.";
     return;
   }
   int found_subgraph_count = 0;

paddle/fluid/framework/ir/seq_concat_fc_fuse_pass.cc

@@ -37,8 +37,8 @@ PDNode* BuildSeqExpandConcatPattern(PDPattern* pattern) {
   // concat output
 
   // So the following variables will be removed:
-  // sequence-expand output
-  // sequence-expand output
+  // sequence_expand output
+  // sequence_expand output
 
   // Three operators
   auto* sequence_expand0 = pattern->NewNode(

paddle/fluid/framework/ir/trt_cross_multihead_matmul_fuse_pass.cc

@@ -311,11 +311,11 @@ int TrtCrossMultiHeadMatmulFusePass::BuildCrossFusion(
     int hidden_out = wq_tensor->dims()[1];
     int head_size = hidden_out / head_number;
     if (abs(scale_attr - 1.0f / sqrt(static_cast<float>(head_size))) > 1e-5) {
-      VLOG(3) << "scale of muilthead matmul do not fit the requirement of "
+      VLOG(3) << "scale of multihead matmul do not fit the requirement of "
                  "flash attention plugin, Stop fusing.";
       return;
     }
-    VLOG(5) << "trt cross attention get wq_tensor name = " << mul0_w->Name()
+    VLOG(5) << "trt cross attention wq_tensor name = " << mul0_w->Name()
             << "trt cross attention wk_tensor name = " << mul1_w->Name()
             << "trt cross attention wv_tensor name = " << mul2_w->Name();
 

paddle/fluid/framework/ir/trt_multihead_matmul_fuse_pass.cc

@@ -62,9 +62,6 @@ static int BuildFusion(Graph* graph, const std::string& name_scope) {
                           Node* scale,
                           Node* scale_out) {
     auto scale_attr = PADDLE_GET_CONST(float, scale->Op()->GetAttr("scale"));
-    // auto scale_bias = PADDLE_GET_CONST(float, scale->Op()->GetAttr("bias"));
-    // bool after_scale =
-    //    PADDLE_GET_CONST(bool, scale->Op()->GetAttr("bias_after_scale"));
 
     // create multihead
     OpDesc multihead_op_desc(mul0->Op()->Block());

paddle/fluid/framework/ir/trt_qk_multihead_matmul_fuse_pass.cc

@@ -228,7 +228,7 @@ PDNode* TrtQKMultiHeadMatmulPattern::operator()() {
       .LinksTo({matmul_qk_out_var});
   scale->LinksFrom({matmul_qk_out_var}).LinksTo({scale_out_var});
   softmax_qk->LinksFrom({scale_out_var}).LinksTo({softmax_qk_out_var});
-  // V  path
+  // V path
   mul2->LinksFrom({input1, mul2_w_var}).LinksTo({mul2_out_var});
   elementwise2->LinksFrom({mul2_out_var, elementwise2_w})
       .LinksTo({elementwise2_out});