Skip to content

Commit 0617335

Browse files
heavengateheavengate
committed
fix format. test=develop
1 parent c68c6d5 commit 0617335

9 files changed

+45
-44
lines changed

‎paddle/fluid/framework/ir/fused_multi_transformer_decoder_pass.cc‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -98,7 +98,7 @@ PDNode* FusedMultiTransformerDecoderPattern::operator()() {
9898
->assert_is_op_output("transpose2")
9999
->AsIntermediate()
100100
->assert_is_op_input("matmul", "X");
101-
101+
102102
// Q path Links
103103
matmul0->LinksFrom({layer_norm_out_var, matmul0_w_var}).LinksTo({matmul0_out_var});
104104
eltadd0->LinksFrom({matmul0_out_var, eltadd0_b_var}).LinksTo({eltadd0_out_var});

‎paddle/fluid/framework/ir/fused_multi_transformer_decoder_pass.h‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -138,7 +138,7 @@ struct FusedMultiTransformerDecoderPattern : public PatternBase {
138138
PATTERN_DECL_NODE(ffn_eltadd1_out);
139139
PATTERN_DECL_NODE(ffn_dropout);
140140
PATTERN_DECL_NODE(ffn_dropout_out);
141-
141+
142142
// output elementwise_add
143143
PATTERN_DECL_NODE(ffn_eltadd_out)
144144
PATTERN_DECL_NODE(ffn_output);
@@ -238,7 +238,7 @@ struct FusedMultiTransformerDecoderFuseQKVPattern : public PatternBase {
238238
PATTERN_DECL_NODE(ffn_eltadd1_out);
239239
PATTERN_DECL_NODE(ffn_dropout);
240240
PATTERN_DECL_NODE(ffn_dropout_out);
241-
241+
242242
// output elementwise_add
243243
PATTERN_DECL_NODE(ffn_eltadd_out)
244244
PATTERN_DECL_NODE(ffn_output);

‎paddle/fluid/framework/ir/fused_multi_transformer_decoder_pass_tester.cc‎

Lines changed: 17 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -78,20 +78,20 @@ TEST(FusedMultiTransformerDecoderPass, basic) {
7878
// (reshape_0) transpose2 -> transpose_0
7979
// (reshape_1) transpose2 -> transpose_1
8080
// (reshape_2) transpose2 -> transpose_2
81-
// (transpose_1) concat -> concat_0
82-
// (transpose_2) concat -> concat_2
83-
// (concat_0) assign -> assign_0
84-
// (concat_1) assign -> assign_2
81+
// (transpose_1) concat -> concat_0
82+
// (transpose_2) concat -> concat_2
83+
// (concat_0) assign -> assign_0
84+
// (concat_1) assign -> assign_2
8585
// (transpose_0, transpose_1) matmul -> matmul_qk
8686
// (matmul_qk, bias_qk) elementwise_add -> eltadd_qk
8787
// (eltadd_qk) softmax -> softmax_qk
88-
// (softmax_qk) dropout -> dropout_qk
88+
// (softmax_qk) dropout -> dropout_qk
8989
// (dropout_qk, transpose_2) matmul_v2 -> matmul_qkv
9090
// (matmul_qkv) transpose -> transpose_qkv
9191
// (transpose_qkv) reshape -> reshape_qkv
9292
// (reshape_qkv) matmul_v2 -> matmul_linear
9393
// (matmul_linear) elementwise_add -> eltadd_linear
94-
// (eltadd_linear) dropout -> dropout_linear
94+
// (eltadd_linear) dropout -> dropout_linear
9595
// (eltadd_out) elementwise_add -> attention_out
9696
//
9797
// (attention_out, scale, bias) layer_norm -> ffn_layer_norm_out
@@ -100,11 +100,11 @@ TEST(FusedMultiTransformerDecoderPass, basic) {
100100
// (ffn_eltadd0) gelu -> ffn_gelu
101101
// (ffn_gelu) matmul_v2 -> ffn_matmul1
102102
// (ffn_matmul1, ffn_bias1) elementwise_add -> ffn_eltadd1
103-
// (ffn_eltadd1) dropout -> ffn_dropout
103+
// (ffn_eltadd1) dropout -> ffn_dropout
104104
// (attention_out, ffn_dropout) elementwise_add -> ffn_output
105105

106106
Layers layers;
107-
// MHA: pre LayerNorm
107+
// MHA: pre LayerNorm
108108
auto* x = layers.data("x", {1, 128, 1024});
109109
auto* ln_scale = layers.data("ln_scale", {1024}, true);
110110
auto* ln_bias = layers.data("ln_bias", {1024}, true);
@@ -229,21 +229,21 @@ TEST(FusedMultiTransformerDecoderFuseQKVPass, basic) {
229229
// (matmul_out0, bias_0) elementwise_add -> eltadd_0
230230
// (eltadd_0) reshape2 -> reshape_0
231231
// (reshape_0) transpose2 -> transpose_0
232-
// (transpose_0) split -> split_q, split_k, split_v
233-
// (split_k) concat -> concat_k
234-
// (split_v) concat -> concat_v
235-
// (concat_k) assign -> assign_k
236-
// (concat_v) assign -> assign_v
232+
// (transpose_0) split -> split_q, split_k, split_v
233+
// (split_k) concat -> concat_k
234+
// (split_v) concat -> concat_v
235+
// (concat_k) assign -> assign_k
236+
// (concat_v) assign -> assign_v
237237
// (split_q, split_k) matmul -> matmul_qk
238238
// (matmul_qk, bias_qk) elementwise_add -> eltadd_qk
239239
// (eltadd_qk) softmax -> softmax_qk
240-
// (softmax_qk) dropout -> dropout_qk
240+
// (softmax_qk) dropout -> dropout_qk
241241
// (dropout_qk, transpose_2) matmul_v2 -> matmul_qkv
242242
// (matmul_qkv) transpose -> transpose_qkv
243243
// (transpose_qkv) reshape -> reshape_qkv
244244
// (reshape_qkv) matmul_v2 -> matmul_linear
245245
// (matmul_linear) elementwise_add -> eltadd_linear
246-
// (eltadd_linear) dropout -> dropout_linear
246+
// (eltadd_linear) dropout -> dropout_linear
247247
// (eltadd_out) elementwise_add -> attention_out
248248
//
249249
// (attention_out, scale, bias) layer_norm -> ffn_layer_norm_out
@@ -252,13 +252,13 @@ TEST(FusedMultiTransformerDecoderFuseQKVPass, basic) {
252252
// (ffn_eltadd0) gelu -> ffn_gelu
253253
// (ffn_gelu) matmul_v2 -> ffn_matmul1
254254
// (ffn_matmul1, ffn_bias1) elementwise_add -> ffn_eltadd1
255-
// (ffn_eltadd1) dropout -> ffn_dropout
255+
// (ffn_eltadd1) dropout -> ffn_dropout
256256
// (attention_out, ffn_dropout) elementwise_add -> ffn_output
257257
//
258258
// (transpose_1, transpose_2) while -> decoder block
259259

260260
Layers layers;
261-
// MHA: pre LayerNorm
261+
// MHA: pre LayerNorm
262262
auto* x = layers.data("x", {1, 128, 1024});
263263
auto* ln_scale = layers.data("ln_scale", {1024}, true);
264264
auto* ln_bias = layers.data("ln_bias", {1024}, true);

‎paddle/fluid/framework/ir/fused_multi_transformer_encoder_pass.cc‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -98,7 +98,7 @@ PDNode* FusedMultiTransformerEncoderPattern::operator()() {
9898
->assert_is_op_output("transpose2")
9999
->AsIntermediate()
100100
->assert_is_op_input("matmul", "X");
101-
101+
102102
// Q path Links
103103
matmul0->LinksFrom({layer_norm_out_var, matmul0_w_var}).LinksTo({matmul0_out_var});
104104
eltadd0->LinksFrom({matmul0_out_var, eltadd0_b_var}).LinksTo({eltadd0_out_var});
@@ -838,7 +838,7 @@ inline void QKVWeightsProcessFuseQKV(Tensor* qkv_w_tensor,
838838
for (int j = 0; j < num_head; j++) {
839839
for (int k = 0; k < dim_head; k++) {
840840
int out_idx = i * num_head * dim_head \
841-
+ j * dim_head + k;
841+
+ j * dim_head + k;
842842
int in_idx = j * 3 * dim_head \
843843
+ i * dim_head + k;
844844
tmp_transpose_b_data[out_idx] = qkv_b_data[in_idx];
@@ -955,7 +955,7 @@ int FusedMultiTransformerEncoderPass::BuildFusion(Graph* graph, const std::strin
955955
fused_multi_transformer_op_desc.SetInput("QKVW", {matmul0_w->Name()});
956956
fused_multi_transformer_op_desc.SetInput("QKVBias", {eltadd0_b->Name()});
957957
fused_multi_transformer_op_desc.SetInput("SrcMask", {eltadd_qk_b->Name()});
958-
958+
959959
// CacheKV input
960960
VarDesc cache_kv_desc("cache_kv" + std::to_string(layer_idx));
961961
// FIXME: only support batch_size = 1, and max_seq_len <= 1024
@@ -1467,7 +1467,7 @@ int FusedMultiTransformerEncoderFuseQKVPass::BuildFusion(Graph* graph, const std
14671467
fused_multi_transformer_op_desc.SetInput("QKVW", {matmul0_w->Name()});
14681468
fused_multi_transformer_op_desc.SetInput("QKVBias", {eltadd0_b->Name()});
14691469
fused_multi_transformer_op_desc.SetInput("SrcMask", {eltadd_qk_b->Name()});
1470-
1470+
14711471
// CacheKV input
14721472
VarDesc cache_kv_desc("cache_kv" + std::to_string(layer_idx));
14731473
// FIXME: only support batch_size = 1, and max_seq_len <= 1024
@@ -1519,7 +1519,7 @@ int FusedMultiTransformerEncoderFuseQKVPass::BuildFusion(Graph* graph, const std
15191519
IR_NODE_LINK_TO(eltadd_qk_b, fused_multi_transformer);
15201520

15211521
IR_NODE_LINK_TO(fused_multi_transformer, ffn_output);
1522-
1522+
15231523
// rewrite while OP input
15241524
// 1. delete k, v
15251525
// 2. delete matmul1/2_w eltadd1/2_w

‎paddle/fluid/framework/ir/fused_multi_transformer_encoder_pass.h‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -129,7 +129,7 @@ struct FusedMultiTransformerEncoderPattern : public PatternBase {
129129
PATTERN_DECL_NODE(ffn_eltadd1_out);
130130
PATTERN_DECL_NODE(ffn_dropout);
131131
PATTERN_DECL_NODE(ffn_dropout_out);
132-
132+
133133
// output elementwise_add
134134
PATTERN_DECL_NODE(ffn_eltadd_out)
135135
PATTERN_DECL_NODE(ffn_output);
@@ -224,7 +224,7 @@ struct FusedMultiTransformerEncoderFuseQKVPattern : public PatternBase {
224224
PATTERN_DECL_NODE(ffn_eltadd1_out);
225225
PATTERN_DECL_NODE(ffn_dropout);
226226
PATTERN_DECL_NODE(ffn_dropout_out);
227-
227+
228228
// output elementwise_add
229229
PATTERN_DECL_NODE(ffn_eltadd_out)
230230
PATTERN_DECL_NODE(ffn_output);

‎paddle/fluid/framework/ir/fused_multi_transformer_encoder_pass_tester.cc‎

Lines changed: 11 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -81,13 +81,13 @@ TEST(FusedMultiTransformerEncoderPass, basic) {
8181
// (transpose_0, transpose_1) matmul -> matmul_qk
8282
// (matmul_qk, bias_qk) elementwise_add -> eltadd_qk
8383
// (eltadd_qk) softmax -> softmax_qk
84-
// (softmax_qk) dropout -> dropout_qk
84+
// (softmax_qk) dropout -> dropout_qk
8585
// (dropout_qk, transpose_2) matmul_v2 -> matmul_qkv
8686
// (matmul_qkv) transpose -> transpose_qkv
8787
// (transpose_qkv) reshape -> reshape_qkv
8888
// (reshape_qkv) matmul_v2 -> matmul_linear
8989
// (matmul_linear) elementwise_add -> eltadd_linear
90-
// (eltadd_linear) dropout -> dropout_linear
90+
// (eltadd_linear) dropout -> dropout_linear
9191
// (eltadd_out) elementwise_add -> attention_out
9292
//
9393
// (attention_out, scale, bias) layer_norm -> ffn_layer_norm_out
@@ -96,13 +96,13 @@ TEST(FusedMultiTransformerEncoderPass, basic) {
9696
// (ffn_eltadd0) gelu -> ffn_gelu
9797
// (ffn_gelu) matmul_v2 -> ffn_matmul1
9898
// (ffn_matmul1, ffn_bias1) elementwise_add -> ffn_eltadd1
99-
// (ffn_eltadd1) dropout -> ffn_dropout
99+
// (ffn_eltadd1) dropout -> ffn_dropout
100100
// (attention_out, ffn_dropout) elementwise_add -> ffn_output
101101
//
102102
// (transpose_1, transpose_2) while -> decoder block
103103

104104
Layers layers;
105-
// MHA: pre LayerNorm
105+
// MHA: pre LayerNorm
106106
auto* x = layers.data("x", {1, 128, 1024});
107107
auto* ln_scale = layers.data("ln_scale", {1024}, true);
108108
auto* ln_bias = layers.data("ln_bias", {1024}, true);
@@ -223,19 +223,19 @@ TEST(FusedMultiTransformerEncoderFuseQKVPass, basic) {
223223
// (matmul_out0, bias_0) elementwise_add -> eltadd_0
224224
// (eltadd_0) reshape2 -> reshape_0
225225
// (reshape_0) transpose2 -> transpose_0
226-
// (transpose_0) split -> split_q, split_k, split_v
227-
// (split_k) assign -> assign_k
228-
// (split_v) assign -> assign_v
226+
// (transpose_0) split -> split_q, split_k, split_v
227+
// (split_k) assign -> assign_k
228+
// (split_v) assign -> assign_v
229229
// (split_q, split_k) matmul -> matmul_qk
230230
// (matmul_qk, bias_qk) elementwise_add -> eltadd_qk
231231
// (eltadd_qk) softmax -> softmax_qk
232-
// (softmax_qk) dropout -> dropout_qk
232+
// (softmax_qk) dropout -> dropout_qk
233233
// (dropout_qk, transpose_2) matmul_v2 -> matmul_qkv
234234
// (matmul_qkv) transpose -> transpose_qkv
235235
// (transpose_qkv) reshape -> reshape_qkv
236236
// (reshape_qkv) matmul_v2 -> matmul_linear
237237
// (matmul_linear) elementwise_add -> eltadd_linear
238-
// (eltadd_linear) dropout -> dropout_linear
238+
// (eltadd_linear) dropout -> dropout_linear
239239
// (eltadd_out) elementwise_add -> attention_out
240240
//
241241
// (attention_out, scale, bias) layer_norm -> ffn_layer_norm_out
@@ -244,13 +244,13 @@ TEST(FusedMultiTransformerEncoderFuseQKVPass, basic) {
244244
// (ffn_eltadd0) gelu -> ffn_gelu
245245
// (ffn_gelu) matmul_v2 -> ffn_matmul1
246246
// (ffn_matmul1, ffn_bias1) elementwise_add -> ffn_eltadd1
247-
// (ffn_eltadd1) dropout -> ffn_dropout
247+
// (ffn_eltadd1) dropout -> ffn_dropout
248248
// (attention_out, ffn_dropout) elementwise_add -> ffn_output
249249
//
250250
// (transpose_1, transpose_2) while -> decoder block
251251

252252
Layers layers;
253-
// MHA: pre LayerNorm
253+
// MHA: pre LayerNorm
254254
auto* x = layers.data("x", {1, 128, 1024});
255255
auto* ln_scale = layers.data("ln_scale", {1024}, true);
256256
auto* ln_bias = layers.data("ln_bias", {1024}, true);

‎paddle/fluid/framework/ir/fused_multi_transformer_pass.cc‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -98,7 +98,7 @@ PDNode* FusedMultiTransformerPattern::operator()() {
9898
->assert_is_op_output("transpose2")
9999
->AsIntermediate()
100100
->assert_is_op_input("matmul", "X");
101-
101+
102102
// Q path Links
103103
matmul0->LinksFrom({layer_norm_out_var, matmul0_w_var}).LinksTo({matmul0_out_var});
104104
eltadd0->LinksFrom({matmul0_out_var, eltadd0_b_var}).LinksTo({eltadd0_out_var});
@@ -566,7 +566,7 @@ int FusedMultiTransformerPass::BuildFusion(Graph* graph, const std::string& name
566566
fused_multi_transformer_op_desc.SetInput("QKVW", {matmul0_w->Name()});
567567
fused_multi_transformer_op_desc.SetInput("QKVBias", {eltadd0_b->Name()});
568568
fused_multi_transformer_op_desc.SetInput("SrcMask", {eltadd_qk_b->Name()});
569-
569+
570570
// CacheKV input
571571
VarDesc cache_kv_desc(
572572
patterns::PDNodeName("cache_kv", layer_norm->Name()));
@@ -631,7 +631,7 @@ int FusedMultiTransformerPass::BuildFusion(Graph* graph, const std::string& name
631631
IR_NODE_LINK_TO(eltadd_qk_b, fused_multi_transformer);
632632

633633
IR_NODE_LINK_TO(fused_multi_transformer, ffn_output);
634-
634+
635635
// // // link CacheKV to while
636636
// // IR_NODE_LINK_TO(cache_kv, while0)
637637
// // unlink origin KV output to while

‎paddle/fluid/framework/ir/fused_multi_transformer_pass.h‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -129,7 +129,7 @@ struct FusedMultiTransformerPattern : public PatternBase {
129129
PATTERN_DECL_NODE(ffn_eltadd1_out);
130130
PATTERN_DECL_NODE(ffn_dropout);
131131
PATTERN_DECL_NODE(ffn_dropout_out);
132-
132+
133133
// output elementwise_add
134134
PATTERN_DECL_NODE(ffn_eltadd_out)
135135
PATTERN_DECL_NODE(ffn_output);

‎paddle/fluid/framework/ir/graph_helper.cc‎

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -767,8 +767,9 @@ void GraphToProgram(const Graph &graph,
767767
block->set_idx(idx);
768768
block->set_parent_idx(kRootBlockIndex);
769769
}
770-
771-
GraphToBlock(*graph.GetSubGraph(idx), block, sort_kind);
770+
771+
GraphToBlock(*graph.GetSubGraph(idx), block, sort_kind,
772+
graph.GetBlockId());
772773
}
773774
} else {
774775
GraphToBlock(graph, block, sort_kind, graph.GetBlockId());

0 commit comments

Comments
 (0)