refine sampling code (#1184)

Author: FrostML

paddlenlp/ops/faster_transformer/transformer/faster_transformer.py

@@ -793,16 +793,16 @@ def sample(self,
         model_inputs = self.prepare_inputs_for_generation(input_ids,
                                                           **model_kwargs)
 
-        if self._decoding_strategy == "sampling" and (top_p == 1.0 and
-                                                      top_k > 0):
-            top_p = 0.0
-        elif self._decoding_strategy == "sampling" and (top_p != 1.0 and
-                                                        top_k == 0):
-            top_k = 0
-        else:
-            raise ValueError(
-                "Only topk sampling or topp sampling are supported. " \
-                "Topk sampling and topp sampling cannot be both applied. ")
+        if self._decoding_strategy == "sampling":
+            if top_p == 1.0 and top_k > 0:
+                top_p = 0.0
+            elif top_p <= 0.0 and top_k == 0:
+                raise ValueError(
+                    "Topk sampling or topp sampling must be applied. " \
+                    "Topk sampling and topp sampling cannot be both applied. ")
+            elif (top_p > 0.0 and top_p < 1.0) and top_k > 0:
+                raise ValueError(
+                    "Topk sampling and topp sampling cannot be both applied. ")
 
         return self.forward(
             model_inputs=model_inputs,
@@ -956,16 +956,16 @@ def sample(self,
         model_inputs = self.prepare_inputs_for_generation(input_ids,
                                                           **model_kwargs)
 
-        if self._decoding_strategy == "sampling" and (top_p == 1.0 and
-                                                      top_k > 0):
-            top_p = 0.0
-        elif self._decoding_strategy == "sampling" and (top_p != 1.0 and
-                                                        top_k == 0):
-            top_k = 0
-        else:
-            raise ValueError(
-                "Only topk sampling or topp sampling are supported. " \
-                "Topk sampling and topp sampling cannot be both applied. ")
+        if self._decoding_strategy == "sampling":
+            if top_p == 1.0 and top_k > 0:
+                top_p = 0.0
+            elif top_p <= 0.0 and top_k == 0:
+                raise ValueError(
+                    "Topk sampling or topp sampling must be applied. " \
+                    "Topk sampling and topp sampling cannot be both applied. ")
+            elif (top_p > 0.0 and top_p < 1.0) and top_k > 0:
+                raise ValueError(
+                    "Topk sampling and topp sampling cannot be both applied. ")
 
         return self.forward(
             model_inputs=model_inputs,