[NeuralChat] fix wrong output of multi-round prediction

intel_extension_for_transformers/neural_chat/models/base_model.py

@@ -391,19 +391,19 @@ def get_conv_template(self, model_path: str, task: str = "") -> Conversation:
         if self.conv_template:
             return
         if not task:
-            self.conv_template = PromptTemplate(self.get_default_conv_template(model_path).name)
+            self.conv_template = PromptTemplate(self.get_default_conv_template(model_path).name, clear_history=True)
         else:
-            clear_after_gen = True
+            clear_history = True
             if task == "completion":
                 name = "alpaca_without_input"
             elif task == "chat":
                 name = "neural-chat-7b-v2"
-                clear_after_gen = False
+                clear_history = False
             elif task == "summarization":
                 name = "summarization"
             else:
                 raise NotImplementedError(f"Unsupported task {task}.")
-            self.conv_template = PromptTemplate(name, clear_after_gen=clear_after_gen)
+            self.conv_template = PromptTemplate(name, clear_history=clear_history)
 
     def prepare_prompt(self, prompt: str, model_path: str, task: str = ""):
         self.get_conv_template(model_path, task)

intel_extension_for_transformers/neural_chat/prompts/prompt.py

@@ -202,9 +202,9 @@
 )
 
 class PromptTemplate:
-    def __init__(self, name="one_shot", clear_after_gen=False):
+    def __init__(self, name="one_shot", clear_history=False):
         self.conv = get_conv_template(name)
-        self.clear_after_gen = clear_after_gen
+        self.clear_history = clear_history
 
     @property
     def roles(self):
@@ -215,7 +215,7 @@ def append_message(self, role: str, message: str):
 
     def get_prompt(self) -> str:
         res = self.conv.get_prompt()
-        if self.clear_after_gen:
+        if self.clear_history:
             self.clear_messages()
         return res