add NPU support for several models (#919)

models/contentunderstanding/tagspace/net.py

@@ -33,14 +33,21 @@ def __init__(self, vocab_text_size, vocab_tag_size, emb_dim, hid_dim,
         self.neg_size = neg_size
         self.text_len = text_len
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         self.text_embedding = paddle.nn.Embedding(
             self.vocab_text_size,
             self.emb_dim,
             padding_idx=75377,
-            sparse=True,
+            sparse=use_sparse,
             name="text_emb")
         self.tag_embedding = paddle.nn.Embedding(
-            self.vocab_tag_size, self.emb_dim, sparse=True, name="tag_emb")
+            self.vocab_tag_size,
+            self.emb_dim,
+            sparse=use_sparse,
+            name="tag_emb")
 
         self.conv = nn.Conv1D(
             in_channels=self.emb_dim,

models/multitask/escm2/net.py

@@ -38,10 +38,14 @@ def __init__(self, sparse_feature_number, sparse_feature_dim, num_field,
             self.gate_num = 2
         self.feature_size = feature_size
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         self.embedding = paddle.nn.Embedding(
             self.sparse_feature_number,
             self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             padding_idx=0,
             weight_attr=paddle.ParamAttr(
                 name="SparseFeatFactors",

models/multitask/esmm/net.py

@@ -28,10 +28,14 @@ def __init__(self, sparse_feature_number, sparse_feature_dim, num_field,
         self.ctr_layer_sizes = ctr_layer_sizes
         self.cvr_layer_sizes = cvr_layer_sizes
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         self.embedding = paddle.nn.Embedding(
             self.sparse_feature_number,
             self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             padding_idx=0,
             weight_attr=paddle.ParamAttr(
                 name="SparseFeatFactors",

models/rank/dcn/net.py

@@ -35,11 +35,15 @@ def __init__(self, sparse_feature_number, sparse_feature_dim,
 
         self.init_value_ = 0.1
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         # sparse coding
         self.embedding = paddle.nn.Embedding(
             self.sparse_feature_number,
             self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             padding_idx=0,
             weight_attr=paddle.ParamAttr(
                 initializer=paddle.nn.initializer.TruncatedNormal(

models/rank/dcn_v2/net.py

@@ -37,11 +37,15 @@ def __init__(self, sparse_feature_number, sparse_feature_dim,
 
         self.init_value_ = 0.1
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         # sparse coding
         self.embedding = paddle.nn.Embedding(
             self.sparse_feature_number,
             self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             padding_idx=0,
             weight_attr=paddle.ParamAttr(
                 initializer=paddle.nn.initializer.TruncatedNormal(

models/rank/deeprec/infer.py

@@ -89,6 +89,7 @@ def main(args):
     # tools.vars
     use_gpu = config.get("runner.use_gpu", True)
     use_xpu = config.get("runner.use_xpu", False)
+    use_npu = config.get("runner.use_npu", False)
     use_visual = config.get("runner.use_visual", False)
     test_data_dir = config.get("runner.test_data_dir", None)
     print_interval = config.get("runner.print_interval", None)
@@ -99,16 +100,20 @@ def main(args):
 
     logger.info("**************common.configs**********")
     logger.info(
-        "use_gpu: {}, use_xpu: {}, use_visual: {}, infer_batch_size: {}, \
+        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_visual: {}, infer_batch_size: {}, \
             test_data_dir: {}, start_epoch: {}, end_epoch: {}, \
                 print_interval: {}, model_load_path: {}"
-        .format(use_gpu, use_xpu, use_visual, infer_batch_size, test_data_dir,
-                start_epoch, end_epoch, print_interval, model_load_path))
+        .format(use_gpu, use_xpu, use_npu, use_visual, infer_batch_size,
+                test_data_dir, start_epoch, end_epoch, print_interval,
+                model_load_path))
     logger.info("**************common.configs**********")
 
     if use_xpu:
         xpu_device = 'xpu:{0}'.format(os.getenv('FLAGS_selected_xpus', 0))
         place = paddle.set_device(xpu_device)
+    elif use_npu:
+        npu_device = 'npu:{0}'.format(os.getenv('FLAGS_selected_npus', 0))
+        place = paddle.set_device(npu_device)
     else:
         place = paddle.set_device('gpu' if use_gpu else 'cpu')
 

models/rank/deeprec/trainer.py

@@ -64,6 +64,7 @@ def main(args):
     # tools.vars
     use_gpu = config.get("runner.use_gpu", True)
     use_xpu = config.get("runner.use_xpu", False)
+    use_npu = config.get("runner.use_npu", False)
     use_visual = config.get("runner.use_visual", False)
     train_data_dir = config.get("runner.train_data_dir", None)
     epochs = config.get("runner.epochs", None)
@@ -75,16 +76,19 @@ def main(args):
 
     logger.info("**************common.configs**********")
     logger.info(
-        "use_gpu: {}, use_xpu: {}, use_visual: {}, train_batch_size: {}, \
+        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_visual: {}, train_batch_size: {}, \
             train_data_dir: {}, epochs: {}, print_interval: {}, \
                 model_save_path: {}"
-        .format(use_gpu, use_xpu, use_visual, train_batch_size, train_data_dir,
-                epochs, print_interval, model_save_path))
+        .format(use_gpu, use_xpu, use_npu, use_visual, train_batch_size,
+                train_data_dir, epochs, print_interval, model_save_path))
     logger.info("**************common.configs**********")
 
     if use_xpu:
         xpu_device = 'xpu:{0}'.format(os.getenv('FLAGS_selected_xpus', 0))
         place = paddle.set_device(xpu_device)
+    elif use_npu:
+        npu_device = 'npu:{0}'.format(os.getenv('FLAGS_selected_npus', 0))
+        place = paddle.set_device(npu_device)
     else:
         place = paddle.set_device('gpu' if use_gpu else 'cpu')
 

models/rank/difm/net.py

@@ -88,17 +88,21 @@ def __init__(self, sparse_field_num, sparse_feature_num,
             units_list=fen_layers_size,
             activation="relu")
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         self.sparse_embedding = paddle.nn.Embedding(
             num_embeddings=self.sparse_feature_num,
             embedding_dim=self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="SparseFeatFactors",
                 initializer=paddle.nn.initializer.Uniform()))
         self.sparse_weight = paddle.nn.Embedding(
             num_embeddings=self.sparse_feature_num,
             embedding_dim=1,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="SparseFeatFactors2",
                 initializer=paddle.nn.initializer.Uniform()))

models/rank/dlrm/net.py

@@ -61,10 +61,14 @@ def __init__(self,
             input_shape=self.concat_size + sparse_feature_dim,
             units_list=top_layer_sizes)
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         self.embedding = paddle.nn.Embedding(
             num_embeddings=self.sparse_feature_number,
             embedding_dim=self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="SparseFeatFactors",
                 initializer=paddle.nn.initializer.TruncatedNormal()))

models/rank/dmr/net.py

@@ -48,34 +48,38 @@ def __init__(self, user_size, cms_segid_size, cms_group_id_size,
         self.other_embedding_size = other_embedding_size
         self.history_length = 50
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         self.uid_embeddings_var = paddle.nn.Embedding(
             self.user_size,
             self.main_embedding_size,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="UidSparseFeatFactors",
                 initializer=nn.initializer.Uniform()))
 
         self.mid_embeddings_var = paddle.nn.Embedding(
             self.adgroup_id_size,
             self.main_embedding_size,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="MidSparseFeatFactors",
                 initializer=nn.initializer.Uniform()))
 
         self.cat_embeddings_var = paddle.nn.Embedding(
             self.cate_size,
             self.main_embedding_size,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="CatSparseFeatFactors",
                 initializer=nn.initializer.Uniform()))
 
         self.brand_embeddings_var = paddle.nn.Embedding(
             self.brand_size,
             self.main_embedding_size,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="BrandSparseFeatFactors",
                 initializer=nn.initializer.Uniform()))
@@ -99,15 +103,15 @@ def __init__(self, user_size, cms_segid_size, cms_group_id_size,
         self.campaign_id_embeddings_var = paddle.nn.Embedding(
             self.campaign_id_size,
             self.main_embedding_size,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="CampSparseFeatFactors",
                 initializer=nn.initializer.Uniform()))
 
         self.customer_embeddings_var = paddle.nn.Embedding(
             self.customer_size,
             self.main_embedding_size,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 name="CustomSparseFeatFactors",
                 initializer=nn.initializer.Uniform()))

models/rank/xdeepfm/net.py

@@ -66,11 +66,15 @@ def __init__(self, sparse_feature_number, sparse_feature_dim,
         self.sparse_num_field = sparse_num_field
         self.init_value_ = 0.1
 
+        use_sparse = True
+        if paddle.is_compiled_with_custom_device('npu'):
+            use_sparse = False
+
         # sparse part coding
         self.embedding_one = paddle.nn.Embedding(
             sparse_feature_number,
             1,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 initializer=paddle.nn.initializer.TruncatedNormal(
                     mean=0.0,
@@ -80,7 +84,7 @@ def __init__(self, sparse_feature_number, sparse_feature_dim,
         self.embedding = paddle.nn.Embedding(
             self.sparse_feature_number,
             self.sparse_feature_dim,
-            sparse=True,
+            sparse=use_sparse,
             weight_attr=paddle.ParamAttr(
                 initializer=paddle.nn.initializer.TruncatedNormal(
                     mean=0.0,

tools/infer.py

@@ -68,7 +68,6 @@ def main(args):
     use_auc = config.get("runner.use_auc", False)
     use_xpu = config.get("runner.use_xpu", False)
     use_npu = config.get("runner.use_npu", False)
-    use_ascend = config.get("runner.use_ascend", False)
     use_visual = config.get("runner.use_visual", False)
     test_data_dir = config.get("runner.test_data_dir", None)
     print_interval = config.get("runner.print_interval", None)
@@ -79,10 +78,10 @@ def main(args):
 
     logger.info("**************common.configs**********")
     logger.info(
-        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_ascend: {}, use_visual: {}, infer_batch_size: {}, test_data_dir: {}, start_epoch: {}, end_epoch: {}, print_interval: {}, model_load_path: {}".
-        format(use_gpu, use_xpu, use_npu, use_ascend, use_visual,
-               infer_batch_size, test_data_dir, start_epoch, end_epoch,
-               print_interval, model_load_path))
+        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_visual: {}, infer_batch_size: {}, test_data_dir: {}, start_epoch: {}, end_epoch: {}, print_interval: {}, model_load_path: {}".
+        format(use_gpu, use_xpu, use_npu, use_visual, infer_batch_size,
+               test_data_dir, start_epoch, end_epoch, print_interval,
+               model_load_path))
     logger.info("**************common.configs**********")
 
     if use_xpu:
@@ -91,10 +90,6 @@ def main(args):
     elif use_npu:
         npu_device = 'npu:{0}'.format(os.getenv('FLAGS_selected_npus', 0))
         place = paddle.set_device(npu_device)
-    elif use_ascend:
-        ascend_device = 'ascend:{0}'.format(
-            os.getenv('FLAGS_selected_ascends', 0))
-        place = paddle.set_device(ascend_device)
     else:
         place = paddle.set_device('gpu' if use_gpu else 'cpu')
 

tools/trainer.py

@@ -67,7 +67,6 @@ def main(args):
     use_auc = config.get("runner.use_auc", False)
     use_npu = config.get("runner.use_npu", False)
     use_xpu = config.get("runner.use_xpu", False)
-    use_ascend = config.get("runner.use_ascend", False)
     use_visual = config.get("runner.use_visual", False)
     train_data_dir = config.get("runner.train_data_dir", None)
     epochs = config.get("runner.epochs", None)
@@ -81,10 +80,9 @@ def main(args):
 
     logger.info("**************common.configs**********")
     logger.info(
-        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_ascend: {}, use_visual: {}, train_batch_size: {}, train_data_dir: {}, epochs: {}, print_interval: {}, model_save_path: {}".
-        format(use_gpu, use_xpu, use_npu, use_ascend, use_visual,
-               train_batch_size, train_data_dir, epochs, print_interval,
-               model_save_path))
+        "use_gpu: {}, use_xpu: {}, use_npu: {}, use_visual: {}, train_batch_size: {}, train_data_dir: {}, epochs: {}, print_interval: {}, model_save_path: {}".
+        format(use_gpu, use_xpu, use_npu, use_visual, train_batch_size,
+               train_data_dir, epochs, print_interval, model_save_path))
     logger.info("**************common.configs**********")
 
     if use_xpu:
@@ -93,10 +91,6 @@ def main(args):
     elif use_npu:
         npu_device = 'npu:{0}'.format(os.getenv('FLAGS_selected_npus', 0))
         place = paddle.set_device(npu_device)
-    elif use_ascend:
-        ascend_device = 'ascend:{0}'.format(
-            os.getenv('FLAGS_selected_ascends', 0))
-        place = paddle.set_device(ascend_device)
     else:
         place = paddle.set_device('gpu' if use_gpu else 'cpu')