Submit add_n categorical test script and json

Author: liuruyan

api/tests/add_n.py

@@ -34,22 +34,16 @@ def build_graph(self, config):
 @benchmark_registry.register("add_n")
 class TorchAddN(PytorchOpBenchmarkBase):
     def build_graph(self, config):
-        input_list = []
-        input_0 = self.variable(
-            name='input_' + str(0),
-            shape=config.inputs_shape[0],
-            dtype=config.inputs_dtype[0])
-        result = input_0
-        input_list.append(input_0)
-        for i in range(1, len(config.inputs_shape)):
+        inputs = []
+        for i in range(len(config.inputs_shape)):
             input_i = self.variable(
                 name='input_' + str(i),
                 shape=config.inputs_shape[i],
                 dtype=config.inputs_dtype[i])
-            result = torch.add(result, input_i)
-            input_list.append(input_i)
-
-        self.feed_list = input_list
+            inputs.append(input_i)
+        inputs = torch.stack(inputs, dim=0)
+        result = torch.sum(inputs, axis=0)
+        self.feed_list = inputs
         self.fetch_list = [result]
 
 

api/tests/categorical.py

@@ -0,0 +1,47 @@
+from common_import import *
+
+
+@benchmark_registry.register("categorical")
+class CategoricalConfig(APIConfig):
+    def __init__(self):
+        super(CategoricalConfig, self).__init__("categorical")
+        self.feed_spec = {"range": [-5, -0.1]}
+
+
+@benchmark_registry.register("categorical")
+class PaddleCategorical(PaddleOpBenchmarkBase):
+    def build_graph(self, config):
+        logits = self.variable(
+            name="logits",
+            shape=config.logits_shape,
+            dtype=config.logits_dtype)
+        result = paddle.distribution.Categorical(logits)
+        counts = result.sample([100])
+        self.feed_list = [logits]
+        self.fetch_list = [counts]
+
+
+@benchmark_registry.register("categorical")
+class TorchCategorical(PytorchOpBenchmarkBase):
+    def build_graph(self, config):
+        logits = self.variable(
+            name="logits",
+            shape=config.logits_shape,
+            dtype=config.logits_dtype)
+        result = torch.distributions.categorical.Categorical(
+            logits=torch.tensor(logits))
+        counts = result.sample([100])
+        self.feed_list = [logits]
+        self.fetch_list = [counts]
+
+
+@benchmark_registry.register("categorical")
+class TFCategoricall(TensorflowOpBenchmarkBase):
+    def build_graph(self, config):
+        logits = self.variable(
+            name='logits',
+            shape=config.logits_shape,
+            dtype=config.logits_dtype)
+        counts = tf.random.categorical(logits, 100)
+        self.feed_list = [logits]
+        self.fetch_list = [counts]

api/tests_v2/configs/add_n.json

@@ -89,4 +89,95 @@
             "type": "list<Variable>"
         }
     }
+}, {
+    "op": "add_n",
+    "param_info": {
+        "inputs": {
+            "inputs0": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs1": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "type": "list<Variable>"
+        }
+    }
+}, {
+    "op": "add_n",
+    "param_info": {
+        "inputs": {
+            "inputs0": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs1": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs2": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs3": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs4": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs5": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs6": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "inputs7": {
+                "dtype": "float16",
+                "shape": "[1L]",
+                "type": "Variable"
+            },
+            "type": "list<Variable>"
+        }
+    }
+}, {
+    "op": "add_n",
+    "param_info": {
+        "inputs": {
+            "inputs0": {
+                "dtype": "float16",
+                "shape": "[-1L, 256L]",
+                "type": "Variable"
+            },
+            "inputs1": {
+                "dtype": "float16",
+                "shape": "[-1L, 256L]",
+                "type": "Variable"
+            },
+            "inputs2": {
+                "dtype": "float16",
+                "shape": "[-1L, 256L]",
+                "type": "Variable"
+            },
+            "inputs3": {
+                "dtype": "float16",
+                "shape": "[-1L, 256L]",
+                "type": "Variable"
+            },
+            "type": "list<Variable>"
+        }
+    }
 }]

api/tests_v2/configs/categorical.json

@@ -0,0 +1,10 @@
+[{
+    "op": "categorical",
+    "param_info": {
+        "logits": {
+            "dtype": "float32",
+            "shape": "[524288L, 23L]",
+            "type": "Variable"
+        }
+    }
+}]